Search

In this paper, we address the problem of automatic image segmentation methods applied to the partial automation of the production process of a traditional Sardinian flatbread called pane Carasau for assuring quality control. The study focuses on one of the most critical activities for obtaining an efficient degree of automation: the estimation of the size and shape of the bread sheets during the production phase, to study the shape variations undergone by the sheet depending on some environmental and production variables. The knowledge can thus be used to create a system capable of predicting the quality of the shape of the dough produced and empower the production process. We implemented an image acquisition system and created an efficient machine learning algorithm, based on support vector machines, for the segmentation and estimation of image measurements for Carasau bread. Experiments demonstrated that the method can successfully achieve accurate segmentation of bread sheets images, ensuring that the dimensions extracted are representative of the sheets coming from the production process. The algorithm proved to be fast and accurate in estimating the size of the bread sheets in various scenarios that occurred over a year of acquisitions. The maximum error committed by the algorithm is equal to the 2.2% of the pixel size in the worst scenario and to 1.2% elsewhere.

With the increased usage of Internet of Things (IoT) devices in recent years, various Machine Learning (ML) algorithms have also developed dramatically for attack detection in this domain. However, the ML models are exposed to different classes of adversarial attacks that aim to fool a model into making an incorrect prediction. For instance, label manipulation or label flipping is an adversarial attack where the adversary attempts to manipulate the label of training data that causes the trained model biased and/or with decreased performance. However, the number of samples to be flipped in this category of attack can be restricted, giving the attacker a limited target selection. Due to the great significance of securing ML models against Adversarial Machine Learning (AML) attacks particularly in the IoT domain, this research presents an extensive review of AML in IoT. Then, a classification of AML attacks is presented based on the literature which sheds light on the future research in this domain. Next, this paper investigates the negative impact levels of applying the malicious label-flipping attacks on IoT data. We devise label-flipping scenarios for training a Support Vector Machine (SVM) model. The experiments demonstrate that the label flipping attacks impact the performance of ML models. These results can lead to designing more effective and powerful attack and defense mechanisms in adversarial settings. Finally, we show the weaknesses of the K-NN defense method against the random label flipping attack., 2022 32nd Conference of Open Innovations Association (FRUCT), November 9-11, 2022, Tampere, Finland, Series: Proceeding of the ... Conference of FRUCT Association

The door system is the core part of the elevator. An accurate diagnosis of the door system can aid engineers in troubleshooting and reduce maintenance costs. However, the research of fault diagnosis based on elevator operation and maintenance data is still in its infancy. With the development of the industrial Internet-of-things, real-time monitoring data of elevator can be collected and used for fault diagnosis modeling. This paper investigates a machine learning-based approach to achieve accurate elevator door fault diagnosis. An experimental study was conducted based on the monitoring data collected from the real-world elevator door system. The experimental results revealed that XGBoost algorithm can accurately identify the fault type of the elevator door.

Computer vision syndrome causes vision problems and discomfort mainly due to dry eye. Several studies show that dry eye in computer users is caused by a reduction in the blink rate and an increase in the prevalence of incomplete blinks. In this context, this article introduces Eye-LRCN, a new eye blink detection method that also evaluates the completeness of the blink. The method is based on a long-term recurrent convolutional network (LRCN), which combines a convolutional neural network (CNN) for feature extraction with a bidirectional recurrent neural network that performs sequence learning and classifies the blinks. A Siamese architecture is used during CNN training to overcome the high-class imbalance present in blink detection and the limited amount of data available to train blink detection models. The method was evaluated on three different tasks: blink detection, blink completeness detection, and eye state detection. We report superior performance to the state-of-the-art methods in blink detection and blink completeness detection, and remarkable results in eye state detection., This work was supported in part by the Portuguese Foundation for Science and Technology (FCT) through the framework of the Strategic Funding under Grant UIDB/04650/2020. The work of Beatriz Remeseiro and Oscar Luaces was supported in part by the Ministry of Science and Innovation, Spain, under Grant PID2019-109238GB-C21.

The aim of the publication is to compare the accuracy, precision, sensitivity and F-measure of machine algorithms trained in the classification of authors of works by English authors and the classification of authors of the same works translated into Bulgarian. The algorithms examined are Multinomial Naive Bayes classifier, Bernoulli Naive Bayes classifier, Support Vector Machines, Random Forest, AdaBoost, Decision Tree and K-Nearest Neighbosr. The research results show that in the English author’s classification with an equal number of works in English, Support Vector Machines and Multinomial Naive Bayes classifier receive the highest values of the studied indicators. In Bulgarian texts, the best results are obtained depending on specific authors

In this paper, a two-stage method is proposed for predicting the catch of skipjack tuna (Katsuwonus pelamis) from a 2D sea environmental pattern. Following the assumption that sea water temperature and sea surface height (SSH) which fishermen often use for finding fishing spots has a correlation with the skipjack tuna catch, a new approach of using Faster R-CNN in object detection is proposed. The proposed method consists of two part. In the first part, taking a sea temperature map as input, Faster R-CNN extracts the candidates of where skipjack tuna would be on the map in order to imitate the behaviors of fishers. In the second part, Support Vector Regression (SVR) estimates the catch amount in each candidate. Fater R-CNN is applied to several sea environmental patterns with three different loss functions and compares each performance. The proposed model is evaluated by comparing the result with average fishers’ ability on the skipjack tuna catches and several criteria for evaluating the proposed model. The results show that the proposed method is able to outperform the average fishers’ ability by an average of 3%.

This study proposes to address the economic significance of unpaid taxes by using an automatic system for predicting a tax default. Too little attention has been paid to tax default prediction in the past. Moreover, existing approaches tend to apply conventional statistical methods rather than advanced data analytic approaches, including state-of-the-art machine learning methods. Therefore, existing studies cannot effectively detect tax default information in real-world financial data because they fail to take into account the appropriate data transformations and nonlinear relationships between early-warning financial indicators and tax default behavior. To overcome these problems, this study applies diverse feature transformation techniques and state-of-the-art machine learning approaches. The proposed prediction system is validated by using a dataset showing tax defaults and non-defaults at Finnish limited liability firms. Our findings provide evidence for a major role of feature transformation, such as logarithmic and square-root transformation, in improving the performance of tax default prediction. We also show that extreme gradient boosting and the systematically developed forest of multiple decision trees outperform other machine learning methods in terms of accuracy and other classification performance measures. We show that the equity ratio, liquidity ratio, and debt-to-sales ratio are the most important indicators of tax defaults for 1-year-ahead predictions. Therefore, this study highlights the essential role of well-designed tax default prediction systems, which require a combination of feature transformation and machine learning methods. The effective implementation of an automatic tax default prediction system has important implications for tax administration and can assist administrators in achieving feasible government expenditure allocations and revenue expansions. Tato studie navrhuje řešit ekonomický význam nezaplacených daní pomocí automatického systému pro predikci nesplácení daní. Predikce daňové neschopnosti byla v minulosti věnována příliš malá pozornost. Stávající přístupy navíc spíše používají konvenční statistické metody než pokročilé přístupy analýzy dat, včetně nejmodernějších metod strojového učení. Stávající studie proto nemohou účinně odhalit informace o daňovém selhání v reálných finančních datech, protože nezohledňují vhodné transformace dat a nelineární vztahy mezi finančními ukazateli včasného varování a chováním při daňovém selhání. K překonání těchto problémů používá tato studie různé techniky transformace atributů a nejmodernější přístupy strojového učení. Navrhovaný predikční systém je ověřen pomocí souboru dat zobrazujícího daňové selhání a neselhání u finských společností s ručením omezeným. Naše zjištění poskytují důkazy o významné roli transformace atributů, jako je logaritmická transformace a transformace s odmocninou, při zlepšování výkonnosti predikce daňového selhání. Ukazujeme také, že extrémní gradientní boosting a systematicky rozvíjený les vícenásobných rozhodovacích stromů překonávají ostatní metody strojového učení z hlediska přesnosti a dalších měřítek klasifikační výkonnosti. Ukazujeme, že poměr vlastního kapitálu, poměr likvidity a poměr dluhu k tržbám jsou nejdůležitějšími ukazateli daňového selhání pro předpověď na 1 rok dopředu. Tato studie proto zdůrazňuje zásadní roli dobře navržených systémů pro predikci daňových selhání, které vyžadují kombinaci transformace atributů a metod strojového učení. Efektivní implementace automatického systému predikce daňového selhání má důležité důsledky pro správu daní a může správcům pomoci při dosahování proveditelných alokací vládních výdajů a expanze příjmů.

In recent years, Ransomware has been a critical threat that attacks smartphones. Ransomware is a kind of malware that blocks the mobile’s system and prevents the user of the infected device from accessing their data until a ransom is paid. Worldwide, Ransomware attacks have led to serious losses for individuals and stakeholders. However, the dramatic increase of Ransomware families makes to the process of identifying them more challenging due to their continuously evolved characteristics. Traditional malware detection methods (e.g., statistical-based prevention methods) fail to combat the evolving Ransomware since they result in a high percentage of false positives. Indeed, developing a non-classical, intelligent technique to safeguarding against Ransomware is of significant importance. This paper introduces a new methodology for the detection of Ransomware that is depending on an evolutionary-based machine learning approach. The binary particle swarm optimization algorithm is utilized for tuning the hyperparameters of the classification algorithm, as well as performing feature selection. The support vector machines (SVM) algorithm is used alongside the synthetic minority oversampling technique (SMOTE) for classification. The utilized dataset is collected from various sources, which consists of 10,153 Android applications, where 500 of them are Ransomware. The performance of the proposed approach SMOTE- $t$ BPSO-SVM achieved merits over traditional machine learning algorithms by having the highest scores in terms of sensitivity, specificity, and g-mean.

Human-vehicle classification plays an important role in advanced driver assistance systems (ADAS). The use of millimeter wave (mmWave) radar sensor in human-vehicle classification algorithms is of great significance since the sensor maintains to be robust in severe weather (e.g. fog, snow, etc.). To improve classification accuracy under complex scenes of autonomous driving, a new mmWave radar point cloud classification algorithm is proposed in this paper, which realizes human-vehicle classification employing a newly proposed point cloud feature vector with eleven dimensions and based on kernel support vector machine (SVM) classifier. To verify the validity and robustness of the proposed feature vector, a 77 GHz radar is used to collect two datasets for static and moving objects, respectively, with each dataset taken for pedestrians and vehicles at different distances and angles. Experimental results show that the proposed algorithm achieves higher classification accuracy than a conventional one based on signal features. For the comparison based on the same number of dimensions, the number of dimensions of the proposed feature vector is decreased by removing the features with low significance. Experimental results verify that the proposed algorithm maintains advantage over the conventional one.

Automatic methods for the detection of pathological voice from healthy speech can be considered as potential clinical tools for medical treatment. This study investigates the effectiveness of glottal source information in the detection of pathological voice by comparing the classical pipeline approach to the end-to-end approach. The traditional pipeline approach consists of a feature extractor and a separate classifier. In the former, two sets of glottal features (computed using the quasi-closed phase glottal inverse filtering method) are used together with the widely used openSMILE features. Using both the glottal and openSMILE features extracted from voice utterances and the corresponding healthy/pathology labels, support vector machine (SVM) classifiers are trained. In building end-to-end systems, both raw speech signals and raw glottal flow waveforms are used to train two deep learning architectures: (1) a combination of convolutional neural network (CNN) and multilayer perceptron (MLP), and (2) a combination of CNN and long short-term memory (LSTM) network. Experiments were carried out using three publicly available databases, including dysarthric (the UA-Speech database and the TORGO database) and dysphonic voices (the UPM database). The performance analysis of the detection system based on the traditional pipeline approach showed best results when the glottal features were combined with the baseline openSMILE features. The results of the end-to-end approach indicated higher accuracies (about 2-3 % improvement in all three databases) when glottal flow was used as the raw time-domain input (87.93 % for UA-Speech, 81.12 % for TORGO and 76.66 % for UPM) compared to using raw speech waveform (85.12 % for UA-Speech, 78.83 % for TORGO and 73.71 % for UPM). The evaluation of both approaches demonstrate that automatic detection of pathological voice from healthy speech benefits from using glottal source information.

Developmental dysphasia, also known as specific language impairment (SLI), is a language disorder in children that involves difficulty in speaking and understanding spoken words. Detecting SLI at an early stage is very important for successful speech therapy in children. In this paper, we propose a novel approach based on glottal source features for detecting children with SLI using the speech signal. The proposed method utilizes time-and frequency-domain glottal parameters, which are extracted from the voice source signal obtained using glottal inverse filtering (GIF). In addition, Mel-frequency cepstral coefficient (MFCC) and openSMILE based acoustic features are also extracted from speech utterances. Two machine learning algorithms, namely, support vector machine (SVM) and feed-forward neural network (FFNN), are trained separately for the MFCC, openSMILE and glottal features. A leave-fourteen-speakers-out cross-validation strategy is used for evaluating the classifiers. The experiments are conducted using the SLI speech corpus launched by the LANNA research group. Experimental results show that the glottal parameters contain significant discriminative information required for identifying children with SLI. Furthermore, the complementary nature of glottal parameters is investigated by independently combining these features with the MFCC and openSMILE acoustic features. The overall results indicate that the glottal features when used in combination with MFCC feature set provides the best performance with the FFNN classifier in the speaker-independent scenario.

The strong airflow generated during the operation of the high-speed train will carry sand and dust, causing high-speed sand and dust to hit the surface of the external equipment of the vehicle body, which will increase the surface roughness of the roof insulators, resulting in the changes of the hydrophobicity, fouling characteristics and insulation performance of the insulator. It poses a hidden danger to the safe and stable operation of the train. In our existing surface roughness testing methods, contact detections and optical stylus methods can cause damage to the insulator surface. Infrared detection methods will be disturbed by leakage current. Most methods need to be disassembled for testing. These methods have certain disadvantages. Therefore, this paper proposes a non-contact detection method for the surface roughness of the roof insulators. Firstly, we extract the image information of the insulator surface by hyperspectral imager, then preprocess the extracted hyperspectral image and use the continuous projection algorithm to reduce the data. Finally, we use the support vector machine to construct the insulator surface roughness discriminant model, and successfully realize the hyperspectral detection method of the surface roughness of the roof insulators, and verify the effectiveness of the proposed method by experiments. As a non-contact detection method, this method can detect the surface roughness of the roof insulators in the non-disassembly conditions, and help the field staffs to grasp the surface roughness of the high-speed train roof insulators in time.

Imbalanced classes are a common problem in machine learning, and the computational costs required for proper resampling increases with the data size. In this study, a simple and effective undersampling method, named particle stacking undersampling (PSU) was proposed. Compared with other competing undersampling methods, PSU can significantly reduce the computational costs, while minimizing information loss to prevent a prediction bias. The performance benchmark applied on 55 binary classification problems indicated that the proposed method not only achieved an enhanced classification performance over other well-known undersampling methods (random undersampling, NearMiss-1, NearMiss-2, cluster centroid, edited nearest neighbor, condensed nearest neighbor, and Tomek Links) but also provided a computational simplicity that can be scalable to large data. Moreover, an experiment verified that two propositions forming the basis of the PSU algorithm can also be applied to other undersampling methods to achieve methodological improvements.

A robust adaptive beamformer based on the support vector machines is proposed to solve the rapid degradation of the performance of minimum variance distortionless response beamformer when the steering vector is mismatched. The proposed beamformer assumes that the angle error range in the steering vector can be estimated; the array only responds within this range, and the array response outside the angle error range is considered as the abnormal response. The proposed method is expressed in the form of support vector regression, and Vapnik's $\varepsilon $ -insensitive quadratic loss function which can improve the robustness of beamformer is used as a regular term to punish the abnormal response. The assumption that the array only responds within a certain angle range reflects the sparsity of support vector regression and improves the ability of the algorithm against steering vector mismatch because the energy conservation theorem holds that the response is large. The regular term ensures that a deep null exists in the direction of interference, the interference can be effectively suppressed. The iterative reweighted least square procedure can converge to the real solution of the SVM regression problem and is easy to calculate; thus, it is used to solve the algorithm proposed in this study. The effectiveness of the algorithm is verified through computer simulations.

One of the most applied materials to manufacture critical components in power plants are martensitic steels due to their high creep and oxidation resistance. In this work, the fracture life of martensitic steels that are designed based on the P92 standard is modeled in order to better understand the relation between its service life and its composition and its thermal treatment. This feature is usually studied by performing creep tests, although carrying out tests of this type are really cost and time consuming. To solve this problem, a multivariate analysis and a training-testing model methodology were performed using a dataset formed by 344 creep tests with the final goal of obtaining a model to predict the fracture life of the material based on several nonlinear techniques like support vector machines and artificial neural networks. Once the models were defined based on predicting with the better generalization capability to cover the whole scenario of the problem, those were compared to determine which one was the most accurate among them. Finally, it was concluded that the model’s performance using the proposed methodology based on artificial neural networks got the most accurate results, achieving low errors of approximately 6.14% when predicting creep behavior under long service times. The authors wish to thank to the Basque Government for its support through grant KK-2019-00033 METALCRO2.

The power grid is a build-up of a mesh of thousands of sensors, embedded devices, and terminal units that communicate over different media. The heterogeneity of modern and legacy equipment calls for attention towards diverse network security measures. The critical infrastructure employs different security measures to detect and prevent adversaries, e.g., through signature-based tools. These approaches lack the potential to identify unknown attacks. Machine learning has the prospective to address novel attack vectors. This paper systematically evaluates the efficacy of learning algorithms from different families for intrusion detection in IEC 60870-5-104 protocol. One-class SVM and k-Nearest Neighbour unsupervised learning models show small potential when being tested on the IEC 104 unseen dataset with Area Under the Curve score 0.64 and 0.59, in the same order; and Matthews Correlation Coefficient value 0.3 and 0.2, respectively. The experimental results suggest little feasibility of the evaluated unsupervised learning approaches for anomaly detection in IEC 104 communication and recommend coupling it with other anomaly detection techniques. © 2021 IEEE.

The agriculture sector in Palestine faces several challenges that affect the quality of crop yields, including plant diseases. Plant diseases may be caused by bacteria, viruses, and fungus, among others. Early detection and classification of these diseases allow farmers to reduce the instances and control the effect that the disease may have on their crops. Therefore, this study utilizes machine learning models for the clustering, prediction, and classification of tomato plant diseases in Palestine. The study used 3000 smartphone digital images of five tomato plant diseases (Alternaria Solani; Botrytis Cinerea; Panonychus Citri; Phytophthora Infestans; Tuta Absoluta) collected from three districts across the West Bank (Tulkarem, Jenin, and Tubas). The machine learning models used image embedding and hierarchical clustering techniques in clustering, and the neural network, random Forest, naïve Bayes, SVM, Decision Tree, and Logistic regression for prediction and classification. The models’ accuracy was validated in reference to a tomato plant diseases database created by plant pathogens experts. The clustering model provided 7 diseases clustering with an accuracy rate of 70%, while the neural network and logistic regression models reported performance accuracies of 70.3% and 68.9% respectively. The findings demonstrate that the proposed approach provides acceptable accuracy rates in the clustering, detection, and classification of tomato plant disease. Thus, the deployment of machine learning techniques in the agriculture sector might help Palestinian farmers better manage and control tomato diseases.

6th International Conference on Computer Science and Engineering, UBMK 2021 -- 15 September 2021 through 17 September 2021 -- -- 176826, Classification problems have an important role in the field of machine learning and data mining. Classification problems are used in different areas such as disease diagnosis, estimation of bank customers, drug studies, sentiment analysis. Many classification algorithms have been developed in the literature and these algorithms have many different parameter inputs. In this study, it is aimed to increase the classification success by using hyperparameter optimization algorithms. K-nearest neighbor, support vector machines, decision tree and gradient boosting classification algorithms were applied to the frequently used 'heart and iris' datasets in the literature. Grid search and random search algorithms, which are hyperparameter optimization algorithms, are applied to these selected classification algorithms. As a result of the experimental studies, it has been observed that the accuracy of all classification algorithms increases when hyperparameter optimization algorithms are applied. The parameter values that give the best results are shown. © 2021 IEEE

1st International Conference on Artificial Intelligence and Data Analytics, CAIDA 2021 -- 6 April 2021 through 7 April 2021 -- The rapid increment in internet usage has also resulted in bulk gerenation of text data. Therefore, investigation of new techniques for automatic classification of textual content is needed as manually managing unstructured text is challenging. The main objective of text classification is to train a model such that it should place an unseen text into correct category. In this study, text classification was performed using the Doc2vec word embedding method on the Turkish Text Classification 3600 (TTC-3600) dataset consisting of Turkish news texts and the BBC-News dataset consisting of English news texts. As the classification method, deep learning-based CNN and traditional machine learning classification methods Gauss Naive Bayes (GNB), Random Forest (RF), Naive Bayes (NB) and Support Vector Machine (SVM) are used. In the proposed model, the highest result was obtained as 94.17% in the Turkish dataset and 96.41% in the English dataset in the classification made with CNN. © 2021 IEEE.

Hydrometallurgy is a metallurgical method for processing complex ores and low-grade ores while reducing environmental pollution. The density of the thickening process in hydrometallurgical production is rather poor, and there are many interference factors, resulting in frequent failures in the density of the thickening process. The main focus of this paper is to propose a method of fault monitoring and diagnosis for the density of the thickening process in hydrometallurgy. First, through the support vector machine (SVM) algorithm, the fault detection model is established to monitor the blockage of the underflow pipeline of the thickener. Second, the fault diagnosis model is established by using the random forest algorithm, and particle swarm optimization is used to optimize the fault diagnosis model. The fault type is judged using the optimized diagnosis model, and the corresponding treatment measures are taken accordingly.

Loan status prediction is an effective tool for investment decisions in peer-to-peer (P2P) lending market. In P2P lending market, most borrowers fulfill the repayment plan; however, some of them fail to pay back their loans. Therefore, an imbalanced classification method can be utilized to discriminate such default borrowers. In this context, the aim of this paper is to propose an investment decision model in P2P lending market which consists of fully paid loans classified via the instance-based entropy fuzzy support vector machine (IEFSVM). IEFSVM is a modified version of the existing entropy fuzzy support vector machine (EFSVM) in terms of an instance-based scheme. IEFSVM can reflect the pattern of nearest neighbors entropy with respect to the change of its size instead of fixing it in unified neighborhood size. Therefore, IEFSVM allows the class change of nearest neighbors in the determination of fuzzy membership. Applying the model to the lending club dataset, we determine loans that are predicted to be fully paid. Then, we also provide a multiple regression model to generate an investment portfolio based on non-default loans that are predicted to yield high returns. Throughout the experiment, the empirical results reveal that IEFSVM outperforms not only EFSVM but also the six other state-of-the-art classifiers including the cost-sensitive adaptive boosting, cost-sensitive random forest, EasyEnsemble, random undersampling boosting, weighted extreme learning machine, and cost-sensitive extreme gradient boosting in terms of loan status classification. Also, the investment performance of the multiple regression model using IEFSVM is higher and more robust than that of two other benchmarks. In this regard, we conclude that the proposed investment model is a decent and practical approach to support decisions in the P2P lending market.

Due to the significant use of Arabic language in social media networks, the demand for Arabic sentiment analysis has increased rapidly. Although, numerous sentiment analysis techniques enable people to obtain valuable insights from the opinions shared on social media. However, these techniques are still in their infancy, and the Arabic sentiment analysis domain lacks a compressive survey. Therefore, this study focused on the various characteristics, State-of-the-Art, and the level of sentiment analysis along with the natural language processing applied in the Arabic sentiment analysis. Furthermore, this study also discussed the sentiment analysis of the modern standards and the dialects of Arabic languages along with various machine learning processes and a few popular algorithms. Moreover, this study adds values by critical analysis of two case studies, which displayed an extensive set of the various research communities in this field of sentiment analysis. Finally, open research challenges are investigated, with a focus on the shortage of lexicons; availability; use of Dialect Arabic (DA); lack of corpora and datasets; right to left reading and compound phrases and idioms.

The classification of fresh fruits according to their visual ripeness is typically a subjective and tedious task; consequently, there is a growing interest in the use of non-contact techniques to automate this process. Machine learning techniques, such as artificial neural networks, support vector machines (SVMs), decision trees, and K-nearest neighbor algorithms, have been successfully applied for classification problems in the literature, particularly for images of fruit. However, the particularities of each classification problem make it difficult, if not impossible, to select a general technique that is applicable to all types of fruit. In this paper, the combinations of four machine learning techniques and three color spaces (RGB, HSV, and L*a*b*) were evaluated with regard to their ability to classify Cape gooseberry fruits. To this end, 925 Cape gooseberry fruit samples were collected, and each fruit was manually classified into one of seven different classes according to its level of ripeness. The color values of each fruit image in the three color spaces and their corresponding ripening stages were organized for training and validation following a fivefold cross-validation strategy in an iterative process repeated 100 times. According to the results, the classification of Cape gooseberry fruits by their ripeness level was sensitive to both the color space and the classification technique used. The models based on the L*a*b* color space and the SVM classifier showed the highest f-measure regardless of the color space, and the principal component analysis combination of color spaces improved the performance of the models at the expense of increased complexity.

Internet of Things (IoT) technology provides a large-scale network for information exchange and communication with big data. Because of the openness of IoT devices in the process of signal transmission, the recognition and access of different IoT devices are directly related to the wide application of its system. The radio frequency fingerprinting (RFF) is a unique characteristic closely related to the hardware of IoT devices themselves, which is difficultly tampered. In this paper, four kinds of RF fingerprint feature extraction algorithms based on statistical features are studied. Robust principle component analysis (RPCA) is used for the dimensionality reduction and the support vector machines (SVM) is used for classification. Through theoretical modeling and experimental verification, the reliability and distinguishability of RFFs are extracted and evaluated, and the classification results are displayed in the real IoT equipment environment.

Although radio frequency ablation is the most effective treatment for atrial fibrillation (AF), there is still a high recurrence rate. The purpose of this paper was to initially assess the probability of the recurrence of AF based on the preoperative body surface potential mapping (BSPM) signals, in other words, to predict the efficiency of ablation and assist physicians in developing more effective treatment options. At present, deep learning methods based on convolutional neural networks (CNNs) do not require complex mathematical abstractions or manual interventions; thus, higher computation efficiency can be obtained in such research. However, the use of the fully connected multi-layer perceptron (MLP) algorithms has shown low classification performance. This paper proposes an improved CNN algorithm (CNN-SVM method) for the recurrence classification in AF patients by combining with the support vector machine (SVM) architecture. The algorithm is validated on the preoperative AF signals of 14 patients for classification. All postoperative patients are followed up for one year; ten of them remain in sinus rhythm, whereas the other four turn back to AF. The ECG data for these patients are obtained through the 128-Lead BSPM system. The results show that the proposed CNN-SVM method can automatically extract the characteristic information through the CNN network. The constructed model ultimately achieved an accuracy of 96%, a sensitivity of 88%, and a specificity of 96%. It is concluded that the CNN-SVM method solves the drawbacks of MLP only for separating linear data. It improves the overall performance of AF recurrence classification, thereby providing a valuable reference for doctors to develop personalized treatment plans.

Functional near-infrared spectroscopy (fNIRS), known as a non-invasive optical neuroimaging technique, is currently used to assess brain dynamics during the performance of complex works and everyday tasks. However, the deep learning approaches to distinguish stress levels based on the changes in hemoglobin concentrations have not yet been extensively investigated. In this paper, we evaluated the efficiencies of advanced methods differentiating the rest and task periods during Stroop task experiments. We first explored that the apparent changes of oxy-hemoglobin and deoxy-hemoglobin concentrations associated with two mental stages did exist across each participant. The preprocessing steps, such as converting raw signals into hemoglobin values and filtering to remove various noises involved in fNIRS signals, were performed to obtain the clean dataset, called non-PCA inputs. Next, we applied the principal component analysis (PCA) algorithm to get PCA-inputs before putting non-PCA inputs into our four classifiers. Then, a novel deep learning-based discrimination framework was studied. The conventional machine learning algorithms, including SVM and AdaBoost, produced the best accuracies of 64.74% ± 1.57% and 71.13% ± 2.96%, respectively. In comparison, the deep learning approaches, including deep belief network and convolutional neural network models, have enabled better classification accuracies of 84.26% ± 2.58% and 72.77% ± 1.92%, respectively.

Existing fine-grained categorization methods predominantly conquer challenges independently, while neglecting the fact that patch proposal and feature extraction can reinforce each other. This necessitates to extract the domain-specific representations and localize key (most discriminative) patches alternately, since implicit to fine-grained specialization is the existence of an entry-category visual shared among all categories. In this study, an increasingly specialized perception convolutional neural network (ISP-CNN) is proposed, focusing on a butterfly domain at sub-species level due to the biosystematics structure. Its pipeline is an coarse-to-fine specialization that hierarchically extracts fine-scale features and proposes distinctive patches at multiple scales. Specifically, the framework consists of two highlights, i.e., hierarchical learning support vector machines (HL-SVMs) and patch proposal sub-networks (PPNs). Depending on the confidences obtained in HL-SVMs, the samples are classified at appropriate subset (i.e., sub-family, genus, and sub-species). Then the PPNs zoom the images to shift the focus on the most representational patches by taking previous predictions of HL-SVMs as a reference, while a finer scale network takes as input an amplified attended region from previous scales with gradual steps. As for self-optimization, ISP-CNN is driven by a patch-level loss and a class-level loss, to mutually learn patch proposals and decisions. For effectness verification, a total of 19,368 lab-made images of butterfly specimens spanning 48 sub-species are utilized as testing samples, while 116,208 augmented images are employed for training. ISP-CNN delivers the better or comparable performance, i.e., validation accuracy achieves 93.67% and testing accuracy achieves 92.13%, which outperforms state-of-the-arts.

To operate wastewater treatment plants (WWTPs) with optimized efficiency, influent conditions (ICs) as initial states of inflow fed to WWTPs were monitored to identify potential anomalies that would trigger adverse events or system crash. To employ voluminous measurements for data-driven decisions, the non-linear, non-Gaussian, non-stationary, auto-correlated, cross-correlated, hetero-skedastic, case-specific nature of multivariate environmental datasets must be considered. This research proposed kernel machine learning models, the kernel principal components analysis based one-class support vector machine (KPCA-OCSVM) with various kernels, to learn anomaly-free training set then classify the testing set. A seven-years multivariate ICs time series was introduced with exploratory analysis performed to reveal temporal behaviors and statistical properties. KPCA with polynomial kernels sufficiently output representative features, based on which OCSVM with Gaussian kernels sensitively and specifically identified anomalies in ICs that were previously omitted by WWTP operators. The proposed kernel algorithms surpassed previous linear PCA-based K-nearest-neighbors models, and improved outcomes with limited increase in computation cost. Without requiring linear, Gaussian, stationary, independent, and homo-skedastic qualities from data, the proposed flexible environmental data science approach could be transferred, rebuilt, and tuned conveniently for ICs from different WWTPs.

Maps of above-ground biomass (AGB) using remote sensing (RS) are valuable to countries like the Philippines for multi-purposes including national greenhouse gas reporting, carbon accounting and even reforestation monitoring. As RS data increase, both optical and radar satellite data have been combined to produce higher quality AGB maps, rather than using individual satellites alone. AGB is then produced after establishing a statistical relationship between combined RS data and plot-based AGB often using machine learning (ML) regression tasks. Here we model a country-wide AGB map of the Philippines while assessing the effects of combining freely and easily accessible satellite data as AGB predictors: Landsat-8 (optical RS), and ALOS-2 PALSAR-2 and Sentinel-1 (radar RS). Using annual composites of each satellite data along with other ecological variables and the National Forest Inventory (NFI) 2014–2015, we trained, cross-validated and compared ML models: Random Forest (RF), Support Vector Machine (SVM) and Neural Network (NN). Model evaluation results were found similar between RF and SVM models i.e., 63 and 65 Mg ha−1 RMSE respectively; while the NN model showed the lowest accuracy i.e., $\text{RMSE} =84$ Mg ha−1, likely a consequence of limited training data. The effect of combining Landsat-8 and ALOS-2 PALSAR-2 was indicated by better AGB predictions in the upland forests, while the inclusion of Sentinel-1 improved the AGB estimates in the agricultural lowland. Using the three satellite data and an RF model not only provides the map with the least bias, but also complements the ridge-to-reef topography of the country where woody vegetations exist from forested mountains down to agricultural lands and mangroves. This study is helpful also to other tropical countries in addressing the problem on expensive forest inventories as well as inaccessible forest lands and conflict areas, while providing spatially explicit and reliable AGB estimates.

ISBN 978-1-7281-7575-1 e-ISBN 978-1-7281-7574-4; International audience; In this paper, we present a combined approach for human localization and classification in Autonomous Train application. Our contribution is threefold. (a) The creation of a new dataset for workers wearing orange vests in a railway environment context. (b) A deep learning supervised YOLO object detector for persons detection combined with a linear SVM (Support Vector Machine) classifier for persons classification into workers wearing orange vests or travelers. (c) A realtime vision-based technique for the environment monitoring in a driverless train application. Experimental results evaluate the parameters of our two stages detection approach and show that our algorithm is robust in detecting and classifying railway workers for a real-time implementation on an embedded system. Our implementation on an embedded system allows a detection with a correct classification rate of 98.5 % of accuracy and a classification time of 1 ms per frame.

The objective of this paper is to compare the accuracy of different kernel functions of the SVM method for text classification. As a basis for the research film reviews are used. The authors try to detect the kernel functions and their parameters to achieve high accuracy in movie reviews classification. The studied kernel functions are: polynomial kernel of degree 2, a linear kernel and a radial base kernel. The achieved accuracy is higher than 83%. The experiments show that the sigmoid radial kernel is an inappropriate choice in text classification.

28th Signal Processing and Communications Applications Conference, SIU 2020 --5 October 2020 through 7 October 2020 -- -- 166413 2-s2.0-85100300725 In the diagnosis of chest diseases, the success of classification with computerized decision assistance systems during auscultation and diagnosis by the specialist is increasing day by day. In this study, 100 respiratory sounds (50 normal and 50 abnormal) sampled at 4kHz, collected by electronic stethoscope for 15 seconds in accordance with routine clinical protocols, were used and no further patient data were used. These sounds were classified by using Support Vector Machines (SVM) and Convolutional Neural Networks (CNN) methods by applying normalization, spectrogram image with STFT, preprocessing such as MFCC and extraction of spectral features. The training set was determined as 80 and the test set as 20, and the success of classification after the study was 67% and 80%, respectively. It has been shown that CNN yields better results than SVM with clinical data with a success rate of 80% and also can be used to classify Ral and Rhonchi from pathological sounds and the highest success rate has been demonstrated by the use of the three layers CNN architecture. © 2020 IEEE.

To detect contaminants accidentally included in packaged foods, food industries use an array of systems ranging from metal detectors to X-ray imagers. Low density plastic or glass contaminants, however, are not easily detected with standard methods. If the dielectric contrast between the packaged food and these contaminants in the microwave spectrum is sensible, Microwave Sensing (MWS) can be used as a contactless detection method, which is particularly useful when the food is already packaged. In this paper we propose using MWS combined with Machine Learning (ML). In particular, we report on experiments we did with packaged cocoa-hazelnut spread and show the accuracy of our approach. We also present an FPGA acceleration that runs the ML processing in real-time so as to keep up with the throughput of a production line.

The purpose of this paper is to compare the accuracy and the execution time of machine learning algorithms for classification of texts, written by Bulgarian authors. The algorithms examined are: Multinomial Naive Bayes classifier, Support Vector Machines, Random Forest and AdaBoost. The results show that the Multinomial Naive Bayes classifier is the most accurate and fastest algorithm for classifying texts by two authors with an equal number of poems in Bulgarian language. The ensemble algorithm AdaBoost is the most accurate for unbalanced data classification. The Support Vector Classification has the highest accuracy. In a classification with an unbalanced set of data, the fastest algorithm is Bernoulli Naive Bayes classifier.

Power system transient stability assessment (TSA) can be represented as a machine learning (ML) binary classification problem. Network measurements data, collected from the distributed phasor measurement units during disturbances, constitute a large and imbalanced data set, on which the ML can be applied in order to learn to recognize the loss-of-stability from the various TSA incidents. This dataset, for actual power networks, contains hundreds of features, many of which can possibly be redundant and/or multi-correlated. This paper proposes an ensemble learning approach to the TSA classification problem, which includes a diverse set of base learners united by a voting ensemble. The imbalanced sample distribution and unequal misclassification costs are considered. Proposed approach also considers a feature selection as a pre-processing step, which is based on the importance analysis from different decision trees based models. Proposed ensemble learning model is applied on the IEEE New England 39-bus test case system. The obtained simulation results corroborate excellent performance and robustness of the proposed approach.

International audience; Recently, several studies have proved that High Frequency Oscillations (HFOs) of [80500] Hz are reliable biomarkers for delineating the epileptogenic zone. The total duration of HFOs is extremely short compared to the entire duration of EEG dataset to be analyzed. Therefore, visual marking of HFOs is timeconsuming and laborious process. In order to promote the clinical use of HFOs oscillations as reliable biomarkers of epileptogenic tissue and to conduct large-scale investigations on cerebral HFOs activities, several automatic detection techniques have been proposed over the past few years. In the present framework, we propose a novel approach for detecting HFOs based on Support Vector Machines (SVM). Our method is subsequently compared with six other methods. HFOs detection performance is evaluated in terms of sensitivity, false discovery rate, area under the ROC curve and execution time. Our results demonstrate that SVM approach yields low false detection (FDR = 6.36%) but, in its current implementation, is moderately sensitive to detect HFOs with a sensitivity of 71.06%. © 2020 IEEE.

Sentiment Analysis is a strand of Natural Language Processing that deals with the emotional polarity a given piece of text has. To gain this understanding from just a string of words, we must first consider a suitable way to break down the text to further classify what each part means. This can be done in a plethora of ways, which mostly stem from the understanding of a classifier. We believe that there is a large amount of information stored in structure-based features within the text, for instance, where the writer may place negation-terms not, neither and how this affects the overall polarity. Similarly, how words in sentences or sections, remote to the current-analysed section, may affect the polarity of said section. A combination of features from both a conventional and a structure-based understanding may also provide us with a larger accuracy in polarity. Therefore, this paper aims to explain both conventional sentiment analysis methods with structure-based methods as well as their practices, advantages and disadvantages concluding with how sentiment analysis can move forward with the appropriation of hybrid methods (methods involving motifs, practices and understandings) from conventional and structure-based methods, for classification.

International audience; Automatic detection of epileptic seizures is a very crucial step for diagnosing patients with drug-resistant epilepsies. If visual analysis of long-term electroencephalographic signals is the most reliable technique, automatic seizures detection can help the physicians in comparing seizures and extracting common patterns. In this paper, a new approach to classify background activity and pre-ictal stereoelectroencephalographic signals is proposed. Linear and nonlinear features are extracted directly from the derived intrinsic mode functions of multivariate empirical mode decomposition technique and the classification is performed using support vector machines. The effectiveness of the proposed approach is evaluated using real datasets. Our results show good performance of the proposed approach since an accuracy of 100% is achieved using the first intrinsic mode function and a window size of 1024 samples. © 2020 IEEE.

GNSS is widely used to provide positions in an absolute reference frame in Unmanned Aerial Vehicles (UAV) and Unmanned Ground Vehicles (UGV), where GNSS is merged with the information provided by other sensors. Even if the main goal of GNSS signal processing is the positioning, multifrequency signals are a rich source of information about the propagation environment surrounding the mobile vehicle. In urban and harsh environment, situational awareness is essential to tailor the operations and take proper countermeasure to harsh propagation conditions. Given this framework the present paper will describe the use of GNSS as signals of opportunity for the characterization of the operative environment by processing the GNSS observables through Machine Learning (ML) algorithms that can be used as efficient features extractors. The paper will present some case studies of operational scenarios for UGVs and for a static monitoring station, showing how through combining DSP techniques with both unsupervised and supervised ML algorithms (K-means classes, Support Vector Machines) it is possible to retrieve the information about the propagation scenario for multipath, interference and atmospheric limitations.

Data on 5244 crystalline compounds from the open AFLOWlib repository are used to build machine learning models, which enable to predict important features of phonon spectrum of a material (Debye temperature and Gruneisen parameter) required for simulation of its lattice properties. We build two types of descriptors: the first one contains data solely on the chemical composition of a compound and the second one incorporates information on the elemental properties of atoms that make up the compound and additionally contains several features regarding its crystal structure. The regression models are built using four popular approaches - gradient boosting (GB), random forests (RF), artificial neural networks (ANN) and support vector machines (SVM). Prior the regression a search for the best values of hyperparameters has been performed for each of the model supplemented with a 5-fold cross validation. We compare prediction accuracy of models based on different methods as well as trained on each of the descriptors. © 2020 IEEE.

This letter presents a black-box modeling approach for the prediction of the spectrum of the maximum currents induced on a generic linear load in an automotive radiated immunity test. The proposed approach relies on a parametric Thevenin-based circuit equivalent built from a limited set of measured or simulated data. The frequency-domain behavior of the equivalent voltage source is provided via a metamodel by combining the support vector machine (SVM) regression with a regularized Fourier kernel with a simple adaptive algorithm. The latter allows defining the minimum number of training samples needed to accurately predict the maximum values of the currents induced on a generic linear load for different azimuth directions of the excitation field. The accuracy and the strength of the proposed approach are demonstrated for an example, by comparing the model predictions with the results of a parametric full-wave electromagnetic simulation.

Power system faults are significant problems in power transmission and distribution. Methods based on relay protection actions and electrical component actions have been put forward in recent years. However, they have deficiencies dealing with power system fault. In this paper, a method for data-based line trip fault prediction in power systems using long short-term memory (LSTM) networks and support vector machine (SVM) is proposed. The temporal features of multisourced data are captured with LSTM networks, which perform well in extracting the features of time series for a long-time span. The strong learning and mining ability of LSTM networks is suitable for a large quantity of time series in power transmission and distribution. SVM, with a strong generalization ability and robustness, is introduced for classification to get the final prediction results. Considering the overfitting problem in fault prediction, layer of dropout and batch normalization are added into the network. The complete network architecture is shown in this paper in detail. The parameters are adjusted to fit the specific situation of the actual power system. The data for experiments are obtained from the Wanjiang substation in the China Southern Power Grid. The real experiments prove the proposed method’s improvements compared with current data mining methods. Concrete analyses of results are elaborated in this paper. A discussion of practical applications is presented to demonstrate the feasibility in real scenarios.

Anemia is a condition that affects over 1.6 billion people worldwide untreated, the disease could lead to increased morbidity and mortality during pregnancy, affecting both the mother and the unborn child. This paper presents the measured dielectric properties of whole blood samples from 176 patients obtained from a hematology clinic; with 80 samples from male patients and 96 samples from female patients. Measurements were performed using a Keysight slim form probe and Keysight network analyzer to obtain the dielectric properties over a wide frequency range (500 MHz-8.5 GHz). A multiple linear regression analysis is performed to identify which components of the blood show the highest correlation with changes in the dielectric properties. Hemoglobin (Hgb) is shown to be the biggest predictor of changes in complex permittivity, demonstrating that permittivity measurements at a single frequency can potentially be used to detect anemia. A support vector machines algorithm is trained and tested to classify between blood samples from healthy patients and blood samples from patients with anemia. The classifier is optimized using a Bayesian-optimization approach during 10-fold cross-validation and then the average performance of the final trained classifier is evaluated through 10-fold testing on unseen data sets. Using a clinical definition of anemia defined as patients having a concentration of Hgb

Robotic soccer games, which have become popular, require timely and precise decision-making in a dynamic environment. To address the problems of complexity in a critical situation, policy improvement in robotic soccer games must occur. This paper proposes an adaptive decision-making method that uses reinforcement learning (RL), and the decision-making system for a robotic soccer game is composed of two subsystems. The first subsystem in the architecture for the proposed method criticizes the situation, and the second subsystem implements decision-making policy. Inspired by the support vector machine (SVM), a situation classification method, which is called an improved SVM, embeds a decision tree structure and simultaneously addresses the problems of a large scale and multiple classifications. When a variety of situations that are collected in the field are classified and congregated into the tree structure, the problem of local strategy selection for each individual class of situations over time is regarded as a RL problem and is solved using a Q-learning method. The results of simulations and experiments demonstrate that the proposed method allows satisfactory decision-making.

Proper tuning of hyper-parameters is essential to the successful application of SVM-classifiers. Several methods have been used for this problem: grid search, random search, estimation of distribution Algorithms (EDAs), bio-inspired metaheuristics, among others. The objective of this paper is to determine the optimal method among those that recently reported good results: Bat algorithm, Firefly algorithm, Fruit-fly optimization algorithm, particle Swarm optimization, Univariate Marginal Distribution Algorithm (UMDA), and Boltzmann-UMDA. The criteria for optimality include measures of effectiveness, generalization, efficiency, and complexity. Experimental results on 15 medical diagnosis problems reveal that EDAs are the optimal strategy under such criteria. Finally, a novel performance index to guide the optimization process, that improves the generalization of the solutions while maintaining their effectiveness, is presented.

As a useful classification method, generalized eigenvalue proximal support vector machine (GEPSVM) is recently studied extensively. However, it may suffer from the sensitivity to outliers, since the L2-norm is used as a measure distance. In this paper, based on the robustness of the L1-norm, we propose an improved robust L1-norm nonparallel proximal SVM with an arbitrary Lp-norm regularization (LpNPSVM), where p > 0. Compared with GEPSVM, the LpNPSVM is more robust to outliers and noise. A simple but effective iterative technique is introduced to solve the LpNPSVM, and its convergence guarantee is also given when 0

Parkinson's disease (PD) is a progressive neurodegenerative motor system disorder. Early diagnosis of PD is important to control the symptoms appropriately. Recent voice and speech recognition techniques provide alternative solutions for PD screening. In this paper, an optimal support vector machine (SVM) based on bacterial foraging optimization (BFO) was established to predict PD effectively. The effectiveness of the proposed method, BFO-SVM, was validated on a PD data set based on vocal measurements. The proposed method was compared with two of the most frequently used parameter optimization methods, including an SVM based on the grid search method and an SVM based on particle swarm optimization. Additionally, to further boost the prediction accuracy, the relief feature selection was employed prior to the BFO-SVM method, consequently the RF-BFO-SVM was proposed. The experimental results have demonstrated that the proposed framework exhibited excellent classification performance with a superior classification accuracy of 97.42%.

An accurate model is important for design and optimization control of centrifugal compressor. However, due to the varying operation conditions and the complexity of the flow dynamics inside it, establishing a satisfaction model of a new compressor is time and cost consuming. This paper proposes a rapid modeling method for centrifugal compressor based on model migration and support vector machines (SVMs). A general framework for centrifugal compressor model migration is given between two similar compressors, the base model of an existing old compressor is revised to fit for the new compressor by SVM using a small number of data. The effective of the proposed method is contrasted with pure SVM by a simulation case and the results show that, compared with the pure SVM, the migrated model can fit the new compressor faster with better accuracy, which is useful for reducing the cost and time in developing the model for new compressors.