Search

The existing broiler health monitoring technology has problems such as low automation, unstable monitoring results, and low practical value, making it difficult to provide timely and reliable broiler health monitoring results. The broiler sound signal can provide feedback on their health. A widely validated and correct experience is to analyze the frequency of coughs in a segment of broiler sound signal to determine the health of the broiler group. Based on this, in this paper, the authors proposed a new broiler health monitoring technology based on sound detection. The broiler health monitoring problem is cleverly transformed into a multi-classification problem, which can be solved by identifying the sound types in broiler sound signals. Specifically, the audio signal collection system was designed to complete signal collection and preliminary signal filtering. Wiener filtering was used for deep signal filtering. The 60-dimensional sound features with good performance from three aspects, time-frequency domain, Mel-Frequency Cepstral Coefficients, and sparse representation were extracted, and a preliminary data set was created. Min-max normalization was used to align the numerical distribution of the data set, and a high-quality data set was created. Multi-classification models based on different classification algorithms and neural networks were trained, and the best-performing Random Forest was obtained, thus parameter optimization was carried out, and the optimal multi-classification model was obtained, achieving a classification accuracy of 91.14%. The visualization platform was built to process the classification results of the multi-classification model, completing majority voting processing and cough rate calculation, thereby achieving broiler health monitoring. In addition, the definitions of cough rate and prediction accuracy were newly proposed. A large number of experiments have verified the feasibility of the broiler health monitoring technology proposed in this paper, with an average prediction accuracy of 98.97% achieved. • Newly propose a complete broiler health monitoring technology based on sound detection. • Transform the broiler health monitoring problem into the sound type identification problem. • Newly propose an index of cough rate to evaluate the health of broiler groups. • Newly propose a data quality improvement scheme. • Obtain the highest prediction accuracy of broiler health monitoring in this field, currently. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel approach for estimating the burned area of forest fires using the TabNet transformer-based architecture. Forest fires pose a significant threat to ecosystems, and accurate estimation of the affected area is essential for effective disaster management and resource allocation. We conducted a comprehensive analysis of various Machine Learning (ML) and Deep Learning (DL) methods, including Random Forest, Neural Networks, Neural Architecture Search (NAS), TabNet with Transformers, and Self-Supervised Learning with Autoencoders, to identify the most accurate and efficient model for area estimation. Our experiments employed a publicly available dataset, UCI Forest Fires, containing a combination of meteorological, geospatial, and categorical data. We implemented a thorough preprocessing pipeline that included handling categorical variables, standardization, and feature engineering. The results demonstrate that TabNet outperforms other methods, achieving state-of-the-art accuracy and generalization in predicting the target variable with a Mean Squared Error (MSE) of 2319 in training and 7781 in testing. [ABSTRACT FROM AUTHOR]

Machine Learning (ML) offers significant potential for quality management in production with predictive analytics. Key aspects to building ML models are the selection and engineering of features from data. They allow the usage of relevant data for training ML models. Using the right features consequently improves the quality of the ML models. However, feature engineering requires knowledge of the data, data preprocessing techniques, algorithms, the domain, and use case. Hence, automatic feature engineering tools have become popular. In this paper, we investigate how domain experts and automatic tools compare for engineering features based on a time series dataset from production. [ABSTRACT FROM AUTHOR]

Solar photovoltaic (PV) energy plays a vital role in global renewable energy generation. Accurate and reliable solar energy forecasting is the key to improving energy scheduling, planning, and intelligent decision-making. However, existing research mainly focuses on short-term solar energy forecasting and lacks exploration of long-term forecasting. To fill the research gap, this paper proposes a complementary fused method using GPU and XGBoost models to improve the performance of long-term solar energy hourly forecasting. The proposed method includes data preprocessing, feature engineering, training of GRU and XGBoost models, and fusion of the forecasting results. Based on historical solar energy, solar irradiance and numerical weather prediction (NWP) data (e.g. temperature, wind direction, wind speed, etc.) and features extended by feature engineering are used as input data to forecast the hourly solar energy production for the next ten days. In order to reduce the risk of overfitting individual models, a simple but effective fusion technique is used to combine the forecasting results of the XGBoost and GRU models to enhance the robustness of the model. In the 2022 Tianchi UNiLAB Algorithm Competition Track 3: Renewable Energy Power Generation Forecast, first place was won by us among 513 participating teams. To validate the scalability of the proposed method, experiments were conducted with various forecasting horizons using the competition dataset and the additional GEFCom2014 dataset. The proposed method demonstrates superior performance when compared with the state-of-the-art long-term series forecasting model DLinear, with MSE and MAE metrics relatively reduced by 28.3% and 17.4% in forecasting the future 150 steps. • Ensembled XGBoost and GRU models for long-term solar energy hourly forecasting. • A well-designed feature engineering is introduced in this study. • Achieved first place in the UNiLAB Algorithm Competition Track 3 competition. [ABSTRACT FROM AUTHOR]

Developing a high-performing traffic incident-duration prediction model is considered a key component for evaluating the impact of these incidents on the roadway network. Various research studies have developed robust incident-duration prediction models. Still, they have faced many issues in providing an accurate prediction result due to the countless data modeling issues, such as complex correlations, highly skewed data distributions, heteroscedasticity, and outliers. This paper investigates the impact of feature optimization (FO) - a relatively new term encompassing two already-known topics: feature engineering (FE) and feature selection (FS) techniques - on the performance of several machine learning models developed for predicting incident durations. The models developed included multivariate linear regression, decision trees, support vector regressors, K-Nearest Neighbors regression, ensembles, and artificial neural networks. Various FO techniques have been used for each model to derive the massive traffic incidents dataset and repeat the prediction process. Our results show that the proposed filtering, wrapper, and embedded FS techniques can successfully reduce the number of features without sacrificing the prediction performance. Using log-normal transformation to deal with continuous data skewness, min-max normalization to deal with data variability, and principal component analysis (PCA) to reform the dataset into a smaller independent feature subset, FE techniques can enhance the accuracy of incident duration estimation over the assessed ML models. The best-performing FE technique was the PCA since performance improvements were observed across all developed ML models. The best-performing FS technique was the Recursive Feature Elimination, outperforming other tested techniques in reducing model complexity while maintaining model accuracy. [ABSTRACT FROM AUTHOR]

Author profiling consists of extracting their demographic and psychographic information by examining their writings. This information can then be used to improve the reader experience and to detect bots or propagators of hoaxes and/or hate speech. Therefore, author profiling can be applied to build more robust and efficient Knowledge-Based Systems for tasks such as content moderation, user profiling, and information retrieval. Author profiling is typically performed automatically as a document classification task. Recently, language models based on transformers have also proven to be quite effective in this task. However, the size and heterogeneity of novel language models, makes it necessary to evaluate them in context. The contributions we make in this paper are four-fold: First, we evaluate which language models are best suited to perform author profiling in Spanish. These experiments include basic, distilled, and multilingual models. Second, we evaluate how feature integration can improve performance for this task. We evaluate two distinct strategies: knowledge integration and ensemble learning. Third, we evaluate the ability of linguistic features to improve the interpretability of the results. Fourth, we evaluate the performance of each language model in terms of memory, training, and inference times. Our results indicate that the use of lightweight models can indeed achieve similar performance to heavy models and that multilingual models are actually less effective than models trained with one language. Finally, we confirm that the best models and strategies for integrating features ultimately depend on the context of the task. • Study of large language models for conducting Author Profiling in Spanish. • Feature integration improves the performance of large language models. • Interpretability of profiles using linguistic features. • Hyperlinks and hashtag are strong features when discerning between bots and humans. • Stylometry is the linguistic category most relevant in Author Profiling. [ABSTRACT FROM AUTHOR]

Lithium-ion batteries age continuously during usage due to their characteristics and the influence of various external factors, but as degradation deepens, it can lead to an apparent decrease in battery safety and reliability. Therefore, predicting remaining useful life from the current to end state and preventing possible dangerous incidents are essential for battery health management. A novel method from the perspective of providing aging features reference and balance prognostics speed and precision is proposed. Firstly, the used dataset is preprocessed in many ways. Then, through feature engineering, 79 features are extracted in the dataset from three perspectives: direct, evolution, statistics, and most representative 16 features are filtered to form the final feature set based on the method of combining discretization and importance. Feature engineering is established to find crucial information within the dataset that correlates highly with remaining life and represents most battery recession paths. Finally, the sparse autoencoder and Transformer integrated approach is proposed to build the life prediction model, and it can learn the temporal relationship between the feature set and remaining cycles quickly and accurately. According to the forecasting results of 62 batteries with different aging conditions, the error can reach 7.43% by only using the feature sequence under 30 cycle lengths. For early life prediction, the error of using the first 100 cycles data is at most 2.6%. • A novel aging feature filtering approach is proposed based on feature engineering. • A prediction model is established by using a sparse autoencoder and Transformer. • More accurate RUL and early life of a battery can be predicted by using fewer cycle data. • After the feature engineering presented in this paper, the final aging feature set selected has strong applicability. [ABSTRACT FROM AUTHOR]

Although black-box models such as ensemble learning models often provide better predictive performance than intrinsic interpretable models such as logistic regression, black-box models are not still applicable due to the lack of interpretability. Recently, there has been an explosion of work on explainable machine learning techniques, which utilize external algorithms or models to explain the behavior of black-box models. However, it is problematic to explain the black-box model behavior because the explanation provided might not reveal the real mechanism or decision process of black-box models. In this study, instead of using explainable machine learning techniques, an automated feature engineering task was formulated to help logistic regression achieve predictive performance comparable to or even better than black-box models while maintaining interpretability. In this paper, an INterpretable Automated Feature ENgineering (INAFEN) framework was designed for logistic regression. This framework automatically transforms the nonlinear relationships between numerical features and labels into linear relationships, conducts feature cross through association rule mining, and distills knowledge from black-box models. A case study was performed on gastric survival prediction to present the rationality of the feature transformations through INAFEN and benchmark experiments to show the validity of INAFEN. Experimental results on 10 classification tasks demonstrated that INAFEN achieved an average ranking of 2.60 in area under the ROC curve (AUROC), 3.35 in area under the PR curve (AUROC), 3.70 in F1 score and 3.00 in Brier score (among 13 models), outperforming other interpretable baselines and even black-box models. In addition, the interpretability measurement of INAFEN is significantly better than that of black-box models. • An interpretable automated feature engineering framework to improve logistic regression. • Transform non-linear relationships using decision tree-based feature transformation. • Association rule-based feature cross to create new features containing interaction information. • Knowledge distillation generates soft labels from black-box models to guide linear model training. [ABSTRACT FROM AUTHOR]

To increase the reliability of motor drive system, many fault diagnosis approaches have been reported with regard to three-phase Pulse Width Modulation Voltage Source Inverter (PWM-VSI). Based on feature engineering and deep neural network, this paper proposes a fault diagnosis approach for the VSI in three-phase Permanent-magnet Synchronous Motor (PMSM) drive. The three-phase current signals are used for the fault diagnosis of VSI. 10 typical signal features are extracted from the three-phase current signals and used as the input of deep neural network. To improve the diagnostic performance, the network structure of the presented deep neural network is designed like a pyramid. Experimental results show that the presented method can detect not only the single open-circuit faults but also the double open-circuit faults in power switches, with high diagnostic accuracy (more than 95%). Besides, comparison results show that the presented method has strong generalization performance. This paper provides theoretical guidance for the fault diagnosis of VSI in PMSM. [ABSTRACT FROM AUTHOR]

The increasing amount of available data in production systems is associated with great potential for process optimization. Due to lack of a data analytics methodology and low data quality within production these potentials often remain unused. Therefore, in this paper we present a model for data acquisition and data preparation including feature engineering for characteristic sensor signals of production machines. The model allows the extraction of relevant process information from the signal, which can be used for monitoring, KPI tracking, trend analysis and anomaly detection. The approach is evaluated on an industrial turning process. [ABSTRACT FROM AUTHOR]

This paper reports a new knowledge-driven engineered feature learning approach in response to the Global AI Challenge for Building E&M Facilities held by the Electrical and Mechanical Service Department (EMSD) of the Hong Kong SAR. The results were awarded with a Grand Prize by the competition organizer. A dynamically engineered multi-modal feature learning (DEMMFL) method is proposed for predicting the cooling load of two office buildings. The DEMMFL model is estimated with the Lasso-ridge regression and compared with other well-known methods such as the Lasso. The novel approach applies control system knowledge to engineer useful features and explore load patterns for multi-mode modeling. Deep learning methods including LSTM, GRU, and AutoGluon are implemented for automated machine learning and tested in parallel to compare the performance of the proposed model with existing methods. The proposed model is demonstrated to predict long-term cooling load most accurately using engineered features from weather information only. • A new dynamic feature engineering method for office building cooling load prediction. • Slow and fast dynamic features engineered from available weather data only. • Control-knowledge based feature to learn on-hours dynamic responses. • Multi-modal approach with eight day-type features to learn occupancy loads. • Deep recurrent learning and AutoML methods compared to the proposed method. [ABSTRACT FROM AUTHOR]

In this study, the authors combined the research on loose particle signal and component signal identification with the research on loose particle material identification for the first time, providing comprehensive and reliable loose particle detection results. Based on this, a signal detection and material identification method for loose particles inside sealed relays based on fusion classification model is proposed. Due to the limitations of technical means and confidentiality management, the authors made a real sealed relay sample, and took it as the research object. Through the steps of data acquisition, signal processing, feature engineering, and model training, the dedicated component identification feature library and material identification feature library was constructed, respectively, the component identification model and material identification model based on parameter-optimized SVM with linear kernel and XGBoost was trained, respectively. For the seal relay to be tested, through the steps of data acquisition, signal processing and feature engineering, the data set to be tested was created. The component identification model was used to identify component signals with loose particle signals, and the material identification model was used to identify the materials of loose particles. The majority voting process was used to convert the classification results into identification results, resulting in loose particle detection and material identification results. In addition, the general procedure steps of the proposed method for physical testing were given, and the identification accuracy for device-level loose particle detection was newly proposed. The loose particle testing event containing thirty-seven identification tasks shows that, the achieved identification accuracy was 97.30%, and 92.16% of the average classification accuracy was achieved by the component identification model, 80.41% of the average classification accuracy was achieved by the material identification model. This effectively demonstrates the feasibility and practicality of the proposed method in this paper. It is an important supplement to the loose particle detection research, and provides references for signal detection in similar fields. • First combine the loose particle signal detection and material identification study. • First construct the dedicated feature library and perform feature optimization. • Train the optimal component identification model and material identification model. • Newly add majority voting steps to obtain the identification results. • Newly propose the identification accuracy for device-level loose particle detection. [ABSTRACT FROM AUTHOR]

The Internet of Things (IoT) connects daily use devices to the Internet, such as home appliances, health care equipment, sensors, and industrial devices. Concurrently, numerous cyber-attacks target those objects and their backbone IoT networks consecutively. Therefore, several researchers have adopted Machine Learning (ML) and Deep Learning (DL) algorithms to develop efficient Intrusion Detection Systems (IDSs). However, the restricted resources of IoT devices hinder integrating those systems with those tiny devices. Hence, designing lightweight IDSs gets more interest from researchers to build efficient detection models to discard attacks in IoT networks. To give a holistic insight into this research domain, this paper presents a Systematic Literature Review (SLR) to review and analyse the recent ML and DL techniques to lighten the IDS models for detecting attacks in IoT devices. In addition, the literature studies were retrieved from six scientific databases Google Scholar, Science Direct, IEEE Xplore®, Scopus, Web of Science, and Springer. From 4,703 identified records, 57 studies were adopted based on predesigned research questions and inclusion/exclusion criteria. The study's findings illustrate the most recently used ML and DL mechanisms and feature engineering techniques to lighten the proposed IDS models. It also shows the most attacks detected, datasets used, tools and network simulators employed, and evaluation metrics and parameters. Furthermore, it suggests the research challenges and future direction after discussing the limitations of the currently proposed techniques. This study shows that most selected studies are journal articles published in IEEE Xplore®. Furthermore, the most used feature engineering techniques are filter-based, as they deliver better performance and lightness than the developed models. Most studies use correlation algorithms as a feature selection technique. Finally, the most discussed attack in the selected studies is the DoS attack. [ABSTRACT FROM AUTHOR]

The growing quality of computer graphics makes it difficult to classify real photos from rendered images. At the same time, a high-quality classifier that separates real-world photos from the images created by computer graphics allow solving many applied problems: from automatic evaluation of the quality of the image to detecting manually modified photos. The main directions of research in this area are associated with the identification of distinctive features in the images for use in classification models. In this paper, we provide experimental studies of various classification models and groups of features on the image dataset from Vkontakte social network. We show that the best choice from a variety of features for solving the classification problem is the combination of Haralick features, Bag of Visual Words (BoVW) with Speeded-Up Robust Features (SURF) descriptors and Local Binary Pattern (LBP) histograms. In addition, we show that Gradient Boosting model being trained on the specified combination of feature groups outperform other machine learning models with 90% accuracy. [ABSTRACT FROM AUTHOR]

With the continuous development of machine learning, enterprises using machine learning methods to mine potential data information has become a hot topic in the research of major insurance companies. In this paper, the features of auto insurance data are analyzed, and the most important features affecting auto renewal are mined. The random forest (RF), gradient lifting tree (GBDT) and lifting machine algorithm (LightGBM) are compared. The test results show that: LightGBM model with the best superiority and robustness. Features of car insurance business channel, NCD, car age and new car purchase price have a greater impact on whether to renew insurance or not. [ABSTRACT FROM AUTHOR]

• Feature construction for fire detection based on thermal images. • Characterization of the response of thermal imaging sensors in fire scenarios. • Data-driven fuzzy modeling with high transparency and interpretability. • Low complexity models with low computational load for implementation. Fire outbreaks are a serious risk in campsites due to the surroundings and dynamic environment of these areas. Due to climate change, conditions of high ignition propensity are becoming more frequent, leading to an increased need for the development of alternative fire prevention systems that can mitigate the consequences of fire incidents. In this context, this work explores thermal imaging data for early detection of fire outbreaks, aiming for application in a real context. To that end, experimental trials were conducted in laboratory and at the venue of a summer festival under real operation conditions. The datasets acquired are characterized in detail, and a feature engineering process is devised for the analysis of the response of thermal imaging sensors to a fire ignition. To deal with high-dimensional data, the feature construction method follows a statistical color-based approach, that characterizes the dynamic changes in the data using three features. Subsequently, this paper proposes a fuzzy modeling approach based on these variables, which is transparent to interpretation and enables the assessment of patterns modeled. The performance of the classification algorithm for detection of fire outbreaks is evaluated and framed with works in the state-of-the-art for this application. [ABSTRACT FROM AUTHOR]

Feature engineering (or creation of new features) that strengthens the importance of fundamental variables (e.g., cloud cover, partial shading, solar irradiance) is essential in gaining an accurate PV forecast for different use cases. On-grid and off-grid PV systems for residential prosumers, and larger scale PV systems are investigated from the PV forecast point of view. Therefore, in this paper, we propose a robust and scalable method designed for PV systems with various sizes and connectivity that provides accurate predictions. The originality of our research consists in feature engineering and combining the deterministic and stochastic models into a meta-learning Stacked Ensemble Forecast (SEF) method that is properly adjusted considering the connectivity of the PV system. The predictions of several standout Machine Learning (ML) algorithms are stacked and scaled to create an accurate forecast model. The proposed methodology is tested on three case studies that cover different types of PV systems: small (residential) on-grid, off-grid, and industrial PV power plants. In all cases, the SEF model provides accurate predictions with an R2 between 0.91 and 0.99 and MAPE between 0.1 and 0.29, followed closely by the stochastic models that have an average R2 of 0.85 and MAPE of 0.3. The deterministic model provides less accurate predictions with an average R2 of 0.8 and MAPE of 0.37. Furthermore, RMSE and MAE improved with more than 16% using the SEF model compared to the stochastic models and with more than 45% compared to the deterministic model. • Forecast for on-grid/off-grid PV systems for prosumers and larger scale PV systems. • Robust PV forecasting method applied for six PV systems with various connectivity. • Feature engineering: effective solar irradiance, combination and analytical features. • A stacked forecasting method combining of deterministic and stochastic models. [ABSTRACT FROM AUTHOR]

• A novel swin-based feature engineering model has been presented. • A new CT image dataset was collected. • Swin-PHOG-LPQ is a lightweight model and an advanced signal processing model. • Our proposal attained over 95% classification accuracies. Computed tomography (CT) image classification has been the subject of intense research in the area of biomedical image classification with the objective of developing intelligent disorder detection models. In this paper, we aim to detect three disorders in lung CT images: hemothorax, contusion, and pneumothorax. Deep learning models are particularly effective for computer vision tasks. Thus our second goal is to propose a new hand-modeled image classification model that achieves high performance using the shifted windows (swin) architecture. We collected a new lung CT image dataset containing four classes – hemothorax, contusion, pneumothorax, and control – with 2730 CT images. Our proposed swin architecture-based CT image classification model is designed to extract features from patches using the Pyramidal histogram-oriented gradient (PHOG) and local phase quantization (LPQ) methods for directional and textural features. We utilized an iterative neighborhood component analysis (INCA) feature selector for feature selection and classified the chosen features using the k-nearest neighbors (kNN) classifier with 10-fold cross-validation. Finally, majority voting was employed to obtain the final classification. Our proposed Swin-PHOG-LPQ achieved a classification accuracy of 95.53%. We also evaluated our model on two publicly available CT image datasets and achieved classification accuracies of 95.31% and 97.63%, respectively. The high classification accuracies obtained by our proposed Swin-PHOG-LPQ model demonstrate its efficacy in detecting the three disorders in lung CT images. [ABSTRACT FROM AUTHOR]

Online reviews significantly impact consumers who are purchasing or seeking services via the Internet. Businesses and review platforms need to manage these online reviews to avoid misleading customers through fake ones. This necessitates developing intelligent solutions to detect these fake reviews and prevent their negative impact on businesses and customers. Therefore, many fake review detection models have been proposed to help distinguish fake reviews from genuine ones. However, these techniques depend on a limited perspective of features, mainly review content, to detect fake reviews, leading to poor performance in discovering the new patterns of fake review content and the dynamic behaviour of spammers. Therefore, there is still a need to develop new solutions to detect the new patterns of fake reviews. Hence, this paper proposes an explainable multi-view deep learning model to identify fake reviews based on different feature perspectives and classifiers. The proposed model can extract essential features from different perspectives, including review content, reviewer data, and product description. Moreover, we employ an ensemble approach that combines three popular deep learning algorithms: Bi-LSTM, CNN, and DNN, to enhance the performance of the fake review detection model. The results of two real-life datasets presented demonstrated the efficiency of our proposed model, where it outperformed the state-of-the-art methods with improvements ranging from 1% to 7% in terms of the AUC metric. To provide visibility into the outcomes of our proposed model and demonstrate the trust and transparency in the obtained results, we also offer a comprehensive explanation for our model results using Shapely Additive Explanations (SHAP) method and attention techniques. The experimental results prove that our proposed model can provide reasonable explanations that help users understand why specific reviews are classified as fake. [ABSTRACT FROM AUTHOR]

In the context of energy transition in Germany, precise load forecasting enables reducing the impact of increased volatility in power generation induced by renewable energies. This paper presents a machine learning approach to generate a 15 minutes forecasting model of the electric load for the ETA research factory at TU Darmstadt on a factory level. In the first iteration, a feature selection process was conducted to select significant features for machine learning datasets. The raw data contained 1,554 features from machine tools, technical building equipment, the building itself and external factors like the weather. The second iteration examined the forecasting capabilities of six hyperparameter tuned algorithms on the feature selected datasets. In the third iteration, feature engineering and hyperparameter tuning led to an optimized Gradient Boosting Regression Trees (GBRT) algorithm. The results indicate that the utilized machine learning approach is feasible and creates a precise very short term load forecasting model, depending on the use case of the load forecast. [ABSTRACT FROM AUTHOR]

• Scientific theory of pitting corrosion is firstly integrated into machine learning model. • The prediction model of maximum pitting corrosion depth is established via automatic model generation. • A real-world pitting corrosion dataset is employed to verify the superiority of the proposed model. • The proposed model achieves better accuracy and efficiency than other models. Accurate monitoring of pipeline corrosion is important and necessary not only for the normal operation of oil and gas pipelines but also for the reliable and stable supply of energy. To avoid failures of buried steel pipelines, the precise prediction of maximum pitting corrosion depth should be conducted to prevent accidents. In this paper, an automatic machine learning (AML) based approach is developed to automate the construction of corrosion depth prediction model. The engineering theory and domain knowledge are integrated into feature engineering, which is an important part of the machine learning (ML) modeling process, to overcome the drawback of the conventional modeling method of ML. Subsequently, a novel prediction method, so-called theory-guided AML (Tg-AML) is proposed for the maximum depth prediction of pitting corrosion pipeline. Firstly, several new feature variables are constructed based on the corrosion mechanism (empirical model and interaction between input variables). Then, seven different feature subsets are developed based on correlation analysis. To select the suitable feature subset and verify the superiority of Tg-AML, a real-world pitting corrosion dataset is utilized for performance comparison based on evaluation metrics. After acquiring the suitable feature subset, the maximum pitting corrosion depth prediction model that fits the data and is guided by both engineering theory and domain knowledge is established. The results indicate that the proposed model achieves better accuracy and efficiency than other models, such as neural network, and decision tree, with root mean square error (RMSE) being 0.288, mean absolute error (MAE) being 0.174, confidence index (Cl) being 0.933. [ABSTRACT FROM AUTHOR]

Construction productivity estimation lacks a comprehensive, standard, and task-type-independent framework to generate and serialize Machine Learning (ML) models. This research aims to develop an ML management framework for estimating the work crew productivity (crew outputs over their working hours) by addressing various operation and project types. The framework takes advantage of historical data, including information regarding operations' progress, weather conditions, the number of resources, and their composition in a work crew. Daily work reports are used as a principal source of historical data. Various hyperparameters-tuned ML algorithms are adopted and ranked based on their computational complexity and prediction accuracy. The generated productivity prediction models have the flexibility to be reused for the effective planning of various construction projects. Applying the proposed framework to a case study of nine disciplines provided estimation models with high accuracies. This study also discusses the theoretical and practical implications of the presented model development procedure. • This paper provides a ML-based framework for construction crew productivity prediction. • The framework aims to address various operation and project types using daily work reports. • Factors, such as site conditions, resource quantity and their work crew composition are considered. • Various techniques and hyperparameters-tuned ML algorithms are developed and adopted. • The generated models can be reused for the effective planning of various construction projects. [ABSTRACT FROM AUTHOR]

• This paper reviews building load prediction with machine learning techniques. • Review and technical papers are searched by Sub-keyword Synonym Searching method. • Technical papers are reviewed in terms of application, algorithms, and data. • Primary limitations and gaps are identified; future trends are predicted. • A guidance for future technical paper on building load prediction is proposed. The surge of machine learning and increasing data accessibility in buildings provide great opportunities for applying machine learning to building energy system modeling and analysis. Building load prediction is one of the most critical components for many building control and analytics activities, as well as grid-interactive and energy efficiency building operation. While a large number of research papers exist on the topic of machine-learning-based building load prediction, a comprehensive review from the perspective of machine learning is missing. In this paper, we review the application of machine learning techniques in building load prediction under the organization and logic of the machine learning, which is to perform tasks T using Performance measure P and based on learning from Experience E. Firstly, we review the applications of building load prediction model (task T). Then, we review the modeling algorithms that improve machine learning performance and accuracy (performance P). Throughout the papers, we also review the literature from the data perspective for modeling (experience E), including data engineering from the sensor level to data level, pre-processing, feature extraction and selection. Finally, we conclude with a discussion of well-studied and relatively unexplored fields for future research reference. We also identify the gaps in current machine learning application and predict for future trends and development. [ABSTRACT FROM AUTHOR]

In this paper, we address the problem of safety monitoring in underground coal mines. In particular, we investigate and compare practical methods for the assessment of seismic hazards using analytical models constructed based on sensory data and domain knowledge. For our case study, we use a rich data set collected during a period of over five years from several active Polish coal mines. We focus on comparing the prediction quality between expert methods which serve as a standard in the coal mining industry and state-of-the-art machine learning methods for mining high-dimensional time series data. We describe an international data mining challenge organized to facilitate our study. We also demonstrate a technique which we employed to construct an ensemble of regression models able to outperform other approaches used by participants of the challenge. Finally, we explain how we utilized the data obtained during the competition for the purpose of research on the cold start problem in deploying decision support systems at new mining sites. [ABSTRACT FROM AUTHOR]

Energy consumption prediction is critical to intelligent power dispatching and smart grid optimization. However, the task remains challenging due to big data fluctuation and the low accuracy of a single feature. This paper proposes a novel short-term energy consumption prediction method for electric vehicle charging stations based on attention feature engineering and stacked GRU (Gated Recurrent Unit). First, we select several handcrafted features from historical energy consumption data. Second, we assign weights to these handcrafted features using the attention mechanism. Third, we propose a multi-sequence stacked GRU in parallel architecture as the network core to extract the correlations between the base and handcraft features. We mitigate the data fluctuations by modifying the prediction resolution and evaluate our method by comparing it with RNN, stacked-GRU, and DeepDeff GRU. The experiment results demonstrate that our method outperforms the state-of-the-art method with a prediction accuracy improvement of 27.2%. [Display omitted] • A feature processing method based on feature engineering and attention mechanism is proposed. • A multi-sequence stacking GRU training structure is proposed. • The proposed method was tested on the energy consumption dataset of charging stations and compared with other methods. [ABSTRACT FROM AUTHOR]

• Presented an extensive overview on different stages of electrocardiogram analysis. • Summarized the most notable developments of smart health monitoring systems. • Discussed the latest hardware implementations of real-time ECG monitoring systems. • Identified challenges and limitations for various automatic ECG diagnosis systems. • Outlines the future vision of next generation ECG monitoring systems for healthcare. Over the last decade, cardiovascular diseases (CVD's) are the leading cause of death globally. Early prediction of CVD's can help in reducing the complications of high-risk patients. The electrocardiogram (ECG) is an efficient aiding tool, provides accurate information about various cardiac conditions of the human heart. Evaluation and interpretation of ECG signal has become the major goal in current research to identify and mitigate risky cardiovascular conditions. The ECG signal is efficiently analysed and classified using signal processing techniques, ranging from traditional to machine learning approaches and its subbranches, such as deep learning are used for the early detection and diagnosis of cardiac conditions and arrhythmias. The development of novel types of body sensors increases the need for automated, low-cost, real-time, and efficient ECG monitoring systems that can be used at home or in ambulatory settings. IoT in healthcare industry enables the doctors and specialists to diagnose the patient status remotely in a smart and efficient manner. Motivated with all the research done in this area for early diagnosis of cardiac abnormalities and save the life of high-risk patients, we surveyed numerous research works reported in the literature and provide a glance on several aspects of ECG signal monitoring systems. We first introduce the step-by-step framework of ECG signal analysis starting data acquisition to classification, describes each stage analysis of both traditional and advanced machine learning models that have been reported in the literature. We provide a deep discussion on ECG signal acquisition, pre-recorded clinical ECG data taken from the databases, signal processing and denoising, detection of feature based on feature engineering, and signal classification along with the comparative assessment amongst reviewed studies. This work also presents the detailed analysis of smart health care system that employs portable and wearable devices, as well as the Internet of Things (IoT) and wireless technologies, as an innovative medium for data transmission that allows for remote access to a patient's health status at lower costs and with greater efficiency. Moreover, this work provides an extensive knowledge on various hardware platforms that have been adopted for the development of biomedical processors including microcontrollers, FPGA and ASIC. Additionally, the challenges and limitations are discussed in this research field and directions for future work are suggested. This is the first review paper of its kind to present a comprehensive discussion on all aspects of cardiovascular disease monitoring systems in one place. [ABSTRACT FROM AUTHOR]

Dark web is an obscured part of the Internet, specifically used for sharing exploits, data breaches, and other means of cybercrime. Dark web forums provide opportunities to share such data and exploits and assign user reputation and credibility through participation in discussions and sharing data, exploits, and hacks. Such activities can help develop metrics to enable identification of influential mal-actors facilitating efficient and effective defense against emerging cyber threats, particularly zero-day exploits. This paper proposes an AI-inspired framework to identify influencers on dark web social networks (INSPECT) through intelligent analysis of user-profiles, interactions, and other activities. INSPECT framework leverages Feature Engineering, Social Network Analysis, Semantic Analysis, and K-means clustering and calculates an influencer score representing the users' significance within these forums. INSPECT has been evaluated using CrimeBB dataset comprising user profiles and activities within dark web forums to assess its effectiveness in identifying influential users on the dark web forums. [ABSTRACT FROM AUTHOR]

The building load forecasting with data-driven technologies guides the operation of the energy system. For new buildings or existing buildings with inadequate monitoring systems, transfer learning can be used to improve load forecasting accuracy. However, without an information integrity judgment, inappropriate migration features and incorrect transfer learning models will reduce the effectiveness of transfer learning. To bridge the gap, a data integrity judgment method is proposed in this paper to determine whether building features are absent. A max-relevance min-redundancy feature selection method with diffusion kernel density estimation (DKDE-mRMR) is established to select the key migrated features. With the transfer component analysis method used as the transfer learning model, the load data of three office buildings are taken as minimum feature set, source domain and target domain respectively for case study. The results prove that the CV-RMSE of load forecasting is over 41.0% by using insufficient information, whereas it can be reduced by 21.6% when information migration is deployed. Even if the target building after feature transfer is further missing the electricity consumption information for cooling load forecast, the CV-RMSE can still be maintained below 21.1% with good robustness. [ABSTRACT FROM AUTHOR]

Building occupancy information could aid energy preservation while simultaneously maintaining the end-user comfort level. Energy conservation becomes essential since energy resources are scarce and human dependency on appliances is only exponentially increasing. While intrusive sensors (i.e., cameras and microphones) can raise privacy concerns, this paper presents an innovative non-intrusive occupancy detection approach using environmental sensor data (e.g., temperature, humidity, carbon dioxide (CO 2), and light sensors). The proposed scheme transforms multivariate time-series data into images for better encoding and extracting relevant features. The utilized image transformation method is based on data normalization and matrix conversion. Precisely, by representing time-series in 2D space, an encoding kernel can move in two directions while it can move only in one direction when applied to a 1D signal. Moreover, machine learning (ML) and deep learning (DL) techniques are utilized to classify occupancy patterns. Several simulations are used to evaluate the approach; mainly, we investigated pre-trained and custom convolutional neural network (CNN) models. The latter attained an accuracy of 99.00%. Additionally, pixel data are extracted from the generated images and subjected to traditional ML methods. Throughout the numerous comparison settings, it was observed that the latter strategy provided the optimal balance of 99.42% accuracy performance and minimal training time across the occupancy datasets. • Exhaustive literature review on occupancy detection schemes based on sensor data. • Image transformation technique based on data normalization and matrix conversion. • Assessing a variety of pre-trained and custom CNN models to predict occupancy status. • Pixel extraction to gather values and integrate them with conventional ML algorithms. [ABSTRACT FROM AUTHOR]

• Feature engineering based framework for location-sensitive disaster event detection. • Augmentation of context-free and context-based features with place of occurrence. • Evaluation with unsupervised machine learning algorithm with 09 performance metrics. • AHP-TOPSIS based selection of an efficient machine learning algorithm and feature set. • Location-augmented context-based features outperformed traditional textual features. Disaster event detection aims to identify events like terrorist attacks, fire incidents, stampede incidents, building collapse, etc., reported in the online news articles or social media. Place of occurrence of disaster event is a significant feature associated with events for location-sensitive disaster event detection. Efficient feature selection and their augmentation with location information can contribute towards the evolution of traditional approaches and their adoption for location-sensitive disaster event detection leading to improvement in the overall process as a whole. Since the evaluation of event detection techniques deliberates various intrinsic and extrinsic performance metrics, the decision-making for the selection of feature sets is treated as a Multiple-Criteria Decision Making (MCDM) problem. This paper proposes a framework, GeoClust , that is based on feature engineering of traditional textual features in order to enhance their capability for improved location-sensitive disaster event detection. The framework augments context-free and context-based textual feature sets with feature sets of place of occurrence of the events and evaluates their performance using unsupervised machine learning algorithms for various performance metrics. Finally, the best feature set is selected using AHP-TOPSIS technique of MCDM in order to tune the system for automatic and efficient location-sensitive disaster event detection in real-time. Extensive set of experiments have been performed in order to evaluate the framework on a dataset of online news articles reporting disaster events about terrorist attacks, fire incidents, stampede incidents, building collapse and maoist attacks happened at different locations in India. The results show that the location-augmented feature sets significantly improve performance of location-sensitive disaster event detection as compared with traditional feature sets. The results also demonstrate that the context-based feature sets with location-augmentation are ranked higher than the context-free feature sets in MCDM analysis. [ABSTRACT FROM AUTHOR]

In malicious URLs detection, traditional classifiers are challenged because the data volume is huge, patterns are changing over time, and the correlations among features are complicated. Feature engineering plays an important role in addressing these problems. To better represent the underlying problem and improve the performances of classifiers in identifying malicious URLs, this paper proposed a combination of linear and non-linear space transformation methods. For linear transformation, a two-stage distance metric learning approach was developed: first, singular value decomposition was performed to get an orthogonal space, and then a linear programming was used to solve an optimal distance metric. For nonlinear transformation, we introduced Nyström method for kernel approximation and used the revised distance metric for its radial basis function such that the merits of both linear and non-linear transformations can be utilized. 33,1622 URLs with 62 features were collected to validate the proposed feature engineering methods. The results showed that the proposed methods significantly improved the efficiency and performance of certain classifiers, such as k -Nearest Neighbor, Support Vector Machine, and neural networks. The malicious URLs' identification rate of k -Nearest Neighbor was increased from 68% to 86%, the rate of linear Support Vector Machine was increased from 58% to 81%, and the rate of Multi-Layer Perceptron was increased from 63% to 82%. We also developed a website to demonstrate a malicious URLs detection system which uses the methods proposed in this paper. The system can be accessed at: http://url.jspfans.com. • Linear and non-linear space transformation methods for malicious URLs detection. • 331622 URLs with 62 features were collected to validate the proposed methods. • The proposed methods can improve the efficiency and performance of classifiers. • A website was developed using the proposed methods to predict malicious URLs. [ABSTRACT FROM AUTHOR]

Early diagnosis of psychiatric conditions can be enhanced by taking into account eye movement behavior. However, the implementation of prediction algorithms which are able to assist physicians in the diagnostic is a difficult task. In this paper we propose, for the first time, an automatic approach for classification of multiple psychiatric conditions based on saccades. In particular, the goal is to classify 6 medical conditions: Alcoholism, Alzheimer's disease, opioid dependence (two groups of subjects with measurements respectively taken prior to and after administering synthetic opioid), Parkinson's disease, and Schizophrenia. Our approach integrates different feature spaces corresponding to complementary characterizations of the saccadic behavior. We define a multi-view model of saccades in which the feature representations capture characteristic temporal and amplitude patterns of saccades. Four of the current most advanced classification methods are used to discriminate among the psychiatric conditions and leave-one-out cross-validation is used to evaluate the classifiers. Classification accuracies well above the chance levels are obtained for the different classification tasks investigated. The confusion matrices reveal that it is possible to separate conditions into different groups. We conclude that using relatively simple descriptors of the saccadic behavior it is possible to simultaneously classify among 6 different types of psychiatric conditions. Conceptually, our multi-view classification method excels other approaches that focus on statistical differences in the saccadic behavior of cases and controls because it can be used for predicting unseen cases. Classification integrating different characterizations of the saccades can actually help to predict the conditions of new patients, opening the possibility to integrate automatic analysis of saccades as a practical procedure for differential diagnosis in Psychiatry. [ABSTRACT FROM AUTHOR]

[Display omitted] • Alternative to the timely and costly finite element-based simulations of the thermal history in additive manufacturing. • Generic, physics-based features that performs well on various geometries, deposition patterns, meshes, and other simulation properties. • High precision machine learning predictions of thermal fields in additive manufacturing that achieved mean absolute percentage error below 5%. Additive manufacturing (AM) is an emerging manufacturing technology that constructs complex parts through layer-by-layer deposition. The prediction and control of thermal fields during production of AM parts are of crucial importance as the temperature distribution and gradient dictates the microstructures, properties, and performance. Finite element (FE) analyses are commonly conducted to simulate the thermal history of the AM process, but are known to be costly and time-consuming. This paper aims to address the challenge by presenting the essential components of a generic data-driven control framework. The proposed framework utilizes extremely randomized trees and is trained and tested on datasets generated through FE simulations. The datasets contain generic, engineered features constructed based on the physics of the underlying thermal process. The features are transferable between a wide range of cases and have achieved mean absolute percentage errors (MAPE) below 2.5% for predicting nodal temperature profiles. In addition, predictions of entire simulations with machine learning (ML) models trained on datasets from different cases have been conducted with MAPE below 5%. The results demonstrate the transferability of thermal histories between several geometries and significantly reduce the need for expensive FE simulations. We believe that these findings are an important step towards real-time optimization in AM. [ABSTRACT FROM AUTHOR]

Enhanced geothermal systems can provide a substantial share of the global energy demand. There exist several hurdles in the engineering implementations of such geothermal systems. One such hurdle is the accurate monitoring of the fracture networks created in subsurface through hydraulic stimulation of these systems. Micro seismicity associated with the stimulation is the primary means to locate the event hypocenters for estimating the stimulated rock volume. Existing methods for location the hypocenters are restricted to only the highest amplitude impulsive signals that are simultaneously detected on several sensors. Consequently, a large portion (usually ∼99%) of the measurements are left unused. In this paper, an unsupervised manifold-approximation followed by clustering of 3-component accelerometer data is used to analyze the seismicity recorded on a monitoring well. With this method, a larger portion of the measured signal is used for the monitoring of the hydraulic fracture network. We analyze the EGS Collab experiment 1 microseismic data, recorded at the Sanford Underground Research Facility, South Dakota. Using the data from a single three-component accelerometer, the polarization features viz. Azimuth, incidence, rectilinearity, and planarity are used as inputs for the unsupervised manifold approximation followed by clustering. Our study shows that density-based clusters in the projected 3D space correspond to distinct types of hydraulically fractured zones around the injection point. We show that the temporal evolution of these clusters can be used to track fracture creation and propagation. [ABSTRACT FROM AUTHOR]

A cerebrovascular accident or stroke is the second commonest cause of death in the world. If it is not fatal, it can result in paralysis, sensory impairment and significant disability. Rehabilitation plays an important role to help survivors relearn lost skills and assist them to regain independence and thus ameliorate their quality of life. With the development of technology, researchers have come up with new solutions to assist clinicians in monitoring and assessing their patients; as well as making physiotherapy available to all. The objective of this review is to assess the recent developments made in the field of post-stroke rehabilitation using wearable devices for data collection and machine learning algorithms for the exercises' evaluation. To do so, PRISMA guidelines for systematic reviews were followed. Scopus, Lens, PubMed, ScienceDirect and Microsoft academic were electronically searched. Peer-reviewed papers using sensors in post-stroke rehabilitation were included, for the period between 2015 to August 2021. Thirty-three publications that used wearable sensors for patients' assessment were included. Based on that, we have proposed a taxonomy that divided the assessment systems into three categories namely activity recognition, movement classification, and clinical assessment emulation. Moreover, The most commonly employed sensors as well as the most targeted body–limbs, outcome measures, and study designs are reviewed, in addition to the examination of the machine learning approaches starting from the feature engineering to the classification done. Finally, limitations and potential study directions in the field are presented. [ABSTRACT FROM AUTHOR]

In many practical applications, such as fraud detection, credit risk modeling or medical decision making, classification models for assigning instances to a predefined set of classes are required to be both precise and interpretable. Linear modeling methods such as logistic regression are often adopted since they offer an acceptable balance between precision and interpretability. Linear methods, however, are not well equipped to handle categorical predictors with high cardinality or to exploit nonlinear relations in the data. As a solution, data preprocessing methods such as weight of evidence are typically used for transforming the predictors. The binning procedure that underlies the weight-of-evidence approach, however, has been little researched and typically relies on ad hoc or expert-driven procedures. The objective in this paper, therefore, is to propose a formalized, data-driven and powerful method. To this end, we explore the discretization of continuous variables through the binning of spline functions, which allows for capturing nonlinear effects in predictor variables and yields highly interpretable predictors that take only a small number of discrete values. Moreover, we extend the weight-of-evidence approach and propose to estimate the proportions using shrinkage estimators. Together, this method offers an improved ability to exploit both nonlinear and categorical predictors to achieve increased classification precision while maintaining the interpretability of the resulting model and decreasing the risk of overfitting. We present the results of a series of experiments in fraud detection and credit risk settings, which illustrate the effectiveness of the presented approach. • Data-driven and powerful preprocessing techniques before applying regression. • Capturing nonlinear effects in variables and highly interpretable predictors. • Extended WOE where proportions are estimated using shrinkage estimators. • Continuous variables are discretized through the binning of spline functions. • Applied on fraud detection and credit risk dataset. [ABSTRACT FROM AUTHOR]

Feature selection in real-world data mining problems is essential to make the learning task efficient and more accurate. Identifying the best feature selection algorithm, among the many available, is a complex activity that still relies heavily on human experts or some random trial-and-error procedure. Thus, the automated machine learning community has taken some steps towards the automation of this process. In this paper, we address the metalearning challenge of recommending feature selection algorithms by proposing a novel meta-feature engineering model. Our model considers a broad collection of meta-features that enable the study of the relationship between the dataset properties and the feature selection algorithm performance in terms of several criteria. We arrange the input meta-features into eight categories: (i) simple, (ii) statistical, (iii) information-theoretical, (iv) complexity, (v) landmarking, (vi) based on symbolic models, (vii) based on images, and (viii) based on complex networks (graphs). The target meta-features emerge from a multi-criteria performance measure, based on five individual performance indexes, that assesses feature selection methods grounded in information, distance, dependence, consistency, and precision measures. We evaluate our proposal using a recently developed framework that extracts the input meta-features from 213 benchmark datasets, and ranks the assessed feature selection algorithms, to fill in the target meta-features in meta-bases. This evaluation uses five state-of-the-art classification methods to induce recommendation models from meta-bases: C4.5, Random Forest, XGBoost, ANN, and SVM. The results showed that it is possible to reach an average accuracy of up to 90% applying our meta-feature engineering model. This work is the first to use an extensive empirical evaluation to provide a careful discussion of the strengths and limitations of more than 160 meta-features. These meta-features, while designed to aid the task of feature selection algorithm recommendation, can readily be employed in other metalearning scenarios. Therefore, we believe our findings are a valuable contribution to the fields of automated machine learning and data mining, as well as to the feature extraction and pattern recognition communities. • A novel meta-feature engineering model recommends feature selection algorithms. • The proposal obtains promising results from 213 datasets with hit rates of up to 90%. • Some simple, landmarking, image, and graph-based input meta-features highlighted. • A multi-criteria performance measure rigorously assesses candidate algorithms. • Chains of binary or multiclass classifiers can efficiently rank candidate algorithms. [ABSTRACT FROM AUTHOR]

Bitcoin is a kind of decentralized cryptocurrency and widely used in online payment partially for its anonymity mechanism. The anonymity, however, also attracts the usage of cryptocurrency by criminals in ransomware and money laundering, and limits its further application and development. In this paper, we aim to improve Bitcoin's auditability with de-anonymization. Many previous studies have used heuristic clustering or supervised machine learning to analyze the historical transactions for identifying user behaviors. However, heuristic clustering only considers the topological structure of the transaction graph and ignores the transaction attributes. While supervised learning is usually limited by the size of labeled datasets, resulting in an unsatisfactory accuracy. To resolve the above problems, we propose an A daptive W eighted A ttribute P ropagation enhanced community detection model, named AWAP, which considers both the transaction's topological structure and the transaction attributes. We first parse the transaction data from the public ledger and construct a bipartite graph to describe correlations between addresses and transactions. Then, we use a five-step feature engineering pipeline to extract Bitcoin address attributes and build an attribute graph. Finally, we design an adaptive weighted attribute propagation algorithm running on the attribute graph to classify the Bitcoin addresses and identify user behaviors. Extensive experiments highlight that AWAP model achieves 12% higher accuracy and 25% higher F-score on average, compared to the state-of-the-art Bitcoin address classifiers and other community detection algorithms. To evaluate the effectiveness of AWAP, we also present two case studies on Bitcoin address classification and Bitcoin trace-ability in ransomware. [ABSTRACT FROM AUTHOR]

With the increasing transportation activities, road freight transportation has caused significant impacts on sustainability. The necessity of establishing sustainable road freight transportation plans have emerged for businesses. Therefore, it is important to develop decision support models that can be used by managers for sustainable road freight transportation. The objective of study is to predict the profit margin given to customers for freight trucking in sustainable road transportation. For this purpose, the variables affecting the profit margin to be given to the customers in sustainable road freight transportation are determined in the light of experts and managers. A data framework has been created using three-dimensional sustainability factors and customer-based variables. A machine learning-based methodology is developed. The scenario-based empirical investigation has been performed. Two different streams have been developed. The first stream is based on traditional importance analysis and second stream considers Recency, Frequency and Monetary based feature engineering and Discrete Wavelet Transform for noise reduction. Three different machine learning algorithms which are Random Forest, Robust Regression and XGBoost used for these two streams. Six different scenarios are considered. Finally, the proposed methodology is applied to two sustainable road freight transportation firms in Turkey. It is aimed to achieve the best performance for profit margin prediction analysis along with demonstrating that the proposed approach provides higher evaluation accuracy. The benchmarking results revealed the superiority of the proposed approach, which trained and tested the prediction model using the stream of data coming from a combination of Discrete Wavelet Transform and feature extraction. The benchmarking and comparisons enact that enhanced XGBoost algorithms provides the best prediction result. This paper presents a novel approach to predict customer-based profit margin in sustainable road freight transportation sector by combining different machine-learning methods for the first time. This study also provides useful insights about strategic and sustainable development perspectives to managers. • Predicting the profit margin in sustainable road freight transportation. • Three-dimensional sustainability factors and customer-based variables are used. • Developing a new methodology based on machine learning. • Best performing machine learning algorithm is highlighted. • Provides managerial and sustainable development implications. [ABSTRACT FROM AUTHOR]

• Electricity fraud detection in conventional meters. • Combining ML: supervised and unsupervised algorithms. • Extensive feature engineering. • Multivariate Gaussian distribution for anomaly detection. • Light Gradient Boosting algorithm: performance and tuning. Although fraud in electricity consumption is easier to detect when consumption is recorded hourly by smart meters, in most developing countries, where the propensity for fraud is higher, conventional meters are not yet affordable. Fraud detection is easier with time series data-logging due to the periodicity and variability of consumption that reveals deviations from a regular consumption pattern. In contrast, fraud detection with conventional meters remains a significant challenge because anomalies in consumption are well hidden within the normal consumption of other consumers. In this paper, large datasets regarding consumers and invoice data from Tunisia are combined and investigated with several Machine Learning (ML) classification algorithms, to detect irregularities in electricity consumption. By performing extensive feature engineering, including multivariate Gaussian distribution, the efficiency of ensemble classifiers such as Light Gradient Boosting (LGB) outperforms other algorithms and achieves realistic performance from challenging, unbalanced and uncorrelated input datasets. [ABSTRACT FROM AUTHOR]

Modular construction is an attractive building method due to its advantages over traditional stick-built methods in terms of reduced waste and construction time, more control over resources and environment, and easier implementation of novel techniques and technologies in a controlled factory setting. However, efficient and timely decision-making in modular factories requires spatiotemporal information about the resources regarding their locations and activities which motivates the necessity for an automated activity identification framework. Thus, this paper utilizes sound, a ubiquitous data source present in every modular construction factory, for the automatic identification of commonly performed manual activities such as hammering, nailing, sawing, etc. To develop a robust activity identification model, it is imperative to engineer the appropriate features of the data source (i.e., traits of the signal) that provides a compact yet descriptive representation of the parameterized audio signal based on the nature of the sound, which is very dependent on the application domain. In-depth analysis regarding appropriate features selection and engineering for audio-based activity identification in construction is missing from current research. Thus, this research extensively investigates the effects of various features extracted from four different domains related to audio signals (time-, time-frequency-, cepstral-, and wavelet-domains), in the overall performance of the activity identification model. The effect of these features on activity identification performance was tested by collecting and analyzing audio data generated from manual activities at a modular construction factory. The collected audio signals were first balanced using time-series data augmentation techniques and then used to extract a 318-dimensional feature vector containing 18 different feature sets from the abovementioned four domains. Several sensitivity analyses were performed to optimize the feature space using a feature ranking technique (i.e., Relief algorithm), and the contribution of features in the top feature sets using a support vector machine (SVM). Eventually, a final feature space was designed containing a 130-dimensional feature vector and 0.5-second window size yielding about 97% F-1 score for identifying different activities. The contributions of this study are two-fold: 1. A novel means of automated manual construction activity identification using audio signal is presented; and 2. Foundational knowledge on the selection and optimization of the feature space from four domains is provided for future work in this research field. The result of this study demonstrates the potential of the proposed system to be applied for automated monitoring and data collection in modular construction factory in conjunction with other activity recognition frameworks based on computer vision (CV) and/or inertial measurement units (IMU). • An audio-based activity identification framework for manual activities in modular construction is presented. • Methods for feature extraction, optimization, and synthesis for classifying audio-signals is provided. • A multi-class SVM model with 5-fold cross-validation is used to validate the designed feature space. • Designed feature space and window-size are used to evaluate the SVM classification model to identify different activities. [ABSTRACT FROM AUTHOR]

Data driven material research is a hot topic in the cross field of artificial intelligence and materials science. The core of new material prediction is to find the relationship between material structure and properties. In this research, machine learning will have important advantages and play an important role for materials data. In this paper,we put forward a framework combining feature engineering and linear regression to find the correlation between structure and properties from materials data. High temperature superconductor and double perovskites for solar cells were employed to test the feasibility of the method. In the former, we successfully rebuilt a descriptor (ℓζ)−1 from data mining which is consistent with the theoretical formula. In the latter, as an exploration, we obtain a new descriptor χ b 2 rs x 2 e rs x − 1 from data mining which expresses the heat of formation (ΔH F) in the double perovskite. By our experiment, the method can obtain related expressions of structure-property relationship for material.The results show that the method is a simple yet efficient paradigm to construct the structure-property relationship and provides valuable hints to accelerate the process of materials design. Unlabelled Image • Construct appropriate descriptors with feature preprocessing and feature engineering. • Non-linear combination of descriptors can help build structure-property relationship. • The linear regression model combined with descriptors can be used to fit the structure-property relationship of materials. [ABSTRACT FROM AUTHOR]

• Accurate transferrable machine learning based load forecasting of machine tools. • Automated data preprocessing, feature construction and selection process. • Time lag and moving average feature construction increases forecasting accuracy. • Autocorrelation function is a valuable information source of feature construction. With the ongoing integration of renewable energies into the electrical power grid, industrial energy flexibility gains importance. To enable demand response applications, knowledge about the future energy demand is necessary. This paper presents a machine learning process to forecast the very short term load of two machine tools, which can be utilized as a decision support basis for control schemes and measures to increase energy flexibility and decrease energy cost in manufacturing. The presented process is developed and evaluated on production machines in a research factory. The results indicate that the developed machine learning process is feasible and creates an accurate very short term load forecasting model for different production machines. It can be used as a blueprint to develop load forecasting models for other production machines using the historic load profile and various machine and process data. A combination of time series features and an Artificial Neural Network proves to be the most robust model regarding the presented machine tools with achieved coefficients of determination between 0.57 and 0.64 for a 100 step forecast. Improvements are still needed regarding the forecasting accuracy, especially of load peaks, for which different measures are proposed. [ABSTRACT FROM AUTHOR]

• Compared performances of both complex deep learning and simple statistical learning models with different level of feature engineering in modeling the IoT testbed system. • Demonstrated that feature engineering plays a key role in developing successful data-driven machine learning models for predicting key process information of the IoT testbed. • Demonstrated that IoT sensors have great potential in advancing manufacturing process modeling and monitoring. As IoT-enabled manufacturing is still in its infancy, there are several key research gaps that need to be addressed. These gaps include the understanding of the characteristics of the big data generated from industrial IoT sensors, the challenges they present to process data analytics, as well as the specific opportunities that the IoT big data could bring to advance manufacturing. In this paper, we use an inhouse-developed IoT-enabled manufacturing testbed to study the characteristics of the big data generated from the testbed. Since the quality of the data usually has the most impact on process modeling, data veracity is often the most challenging characteristic of big data. To address that, we explore the role of feature engineering in developing effective machine learning models for predicting key process variables. We compare complex deep learning approaches to a simple statistical learning approach, with different level or extent of feature engineering, to explore their pros and cons for potential industrial IoT-enabled manufacturing applications. [ABSTRACT FROM AUTHOR]