Search

For the past few decades, thoracic disease classification using Chest X-Ray images has remained challenging in Artificial Intelligence (AI). The traditional Deep CNNs (DCNNs) always predict with overconfidence, so the accuracy of the architectures is likely to be lower than predictive scores. Also, DCNNs suffer from overfitting due to the large number of parameters, and they require immense data augmentation to reduce overfitting. To overcome these problems, this research implements ResNet-152 with a Mix-Up based regularization method to effectively classify multiclass Chest X-Ray images of the Chest X-ray14 dataset first time. It contains 112,120 Chest X-Ray images across 30,805 patients, with 14 common thoracic disease labels plus one Normal (Healthy) label. The Mix-Up technique creates new samples as a convex combination of training points and disease labels. This Mix-Up-based regularization trains a model by blending image pair and their associated labels. We found that Mix-Up is successful in improving the accuracy of the ResNet-152 model by reducing its overconfidence up to 0.002 and increasing average AUC to 99.2% across all 14 disease classes. The proposed model performed better than ResNet-152 with other regularization methods such as Cut-Out, Cut-Mix, and Aug-Mix Regularization. Compared to the other existing models such as CNN-ELM, PCSANet, and GWSA &LCD models, this model achieved higher average AUC values across all 14 pathologies. [ABSTRACT FROM AUTHOR]

Computer-aided diagnosis (CAD) techniques, exemplified by chest x-ray (CXR)-based methods, offer a cost-effective alternative for early-stage COVID-19 diagnosis compared to expensive options such as polymerase chain reaction (PCR) and computed tomography (CT) scan. Despite efforts to diagnose COVID-19 with CXR-based methods, their performance could be improved by considering the spatial relationships between regions of interest (ROIs) in CXR images. This oversight hinders the ability to accurately identify areas of the human lung most vulnerable to COVID-19. This model implements a two-way classification system to differentiate between lung X-ray impressions, accurately determining whether they are affected or normal. The effectiveness of this system is assessed using metrics such as accuracy, recall, precision, and F1-score. We employed over 2409 samples of X-ray images in the COVID-19 diagnosis process. The results obtained from the VGG16 model showcase outstanding performance, with a recognition rate of 99.58% for X-ray images and 94.29% for CT-scan pictures within the given sample size and two-class categorization. This model surpasses all existing approaches documented in the literature. Medical professionals and healthcare workers can effectively utilize this proposed system, leveraging X-rays and CT scans of human lungs to identify COVID-19 cases accurately. [ABSTRACT FROM AUTHOR]

In the last few years, the use of convolutional neural networks (CNNs) in intrusion detection domains has attracted more and more attention. However, their results in this domain have not lived up to expectations compared to the results obtained in other domains, such as image classification and video analysis. This is mainly due to the datasets used, which contain preprocessed features that are not compatible with convolutional neural networks, as they do not allow a full exploit of all the information embedded in the original network traffic. With the aim of overcoming these issues, we propose in this paper a new efficient convolutional neural network model for network intrusion detection based on raw traffic data (pcap files) rather than preprocessed data stored in CSV files. The novelty of this paper lies in the proposal of a new method for adapting the raw network traffic data to the most suitable format for CNN models, which allows us to fully exploit the strengths of CNNs in terms of pattern recognition and spatial analysis, leading to more accurate and effective results. Additionally, to further improve its detection performance, the structure and hyperparameters of our proposed CNN-based model are automatically adjusted using the self-adaptive differential evolution (SADE) metaheuristic, in which symmetry plays an essential role in balancing the different phases of the algorithm, so that each phase can contribute in an equal and efficient way to finding optimal solutions. This helps to make the overall performance more robust and efficient when solving optimization problems. The experimental results on three datasets, KDD-99, UNSW-NB15, and CIC-IDS2017, show a strong symmetry between the frequency values implemented in the images built for each network traffic and the different attack classes. This was confirmed by a good predictive accuracy that goes well beyond similar competing models in the literature. [ABSTRACT FROM AUTHOR]

We study the multiclass classification problem where the features come from a mixture of time‐homogeneous diffusions. Specifically, the classes are discriminated by their drift functions while the diffusion coefficient is common to all classes and unknown. In this framework, we build a plug‐in classifier which relies on nonparametric estimators of the drift and diffusion functions. We first establish the consistency of our classification procedure under mild assumptions and then provide rates of convergence under different set of assumptions. Finally, a numerical study supports our theoretical findings. [ABSTRACT FROM AUTHOR]

For high-dimensional classification, interpolation of training data manifests as the data piling phenomenon, in which linear projections of data vectors from each class collapse to a single value. Recent research has revealed an additional phenomenon known as the 'second data piling' for independent test data in binary classification, providing a theoretical understanding of asymptotically perfect classification. This paper extends these findings to multi-category classification and provides a comprehensive characterization of the double data piling phenomenon. We define the maximal data piling subspace, which maximizes the sum of pairwise distances between piles of training data in multi-category classification. Furthermore, we show that a second data piling subspace that induces data piling for independent data exists and can be consistently estimated by projecting the negatively-ridged discriminant subspace onto an estimated 'signal' subspace. By leveraging this second data piling phenomenon, we propose a bias-correction strategy for class assignments, which asymptotically achieves perfect classification. The present research sheds light on benign overfitting and enhances the understanding of perfect multi-category classification of high-dimensional discrimination with a help of high-dimensional asymptotics. [ABSTRACT FROM AUTHOR]

This paper describes a method for generating realistic synthetic images using a generative deep learning model. When generating training examples, the features of climate, relief, architecture, and landscape of a particular area are considered. The generative model is represented by a variational autoencoder consisting of an encoder, a latent space, and a decoder. To train the autoencoder, a new dataset CSCDroneCL_v1 was created, consisting of manually labeled panoramic images and training examples selected from the open VisDrone2022 dataset. Based on the SOTA-ConvNet template, a discriminative neural network model was designed and trained to classify anomalies obtained from multitemporal panoramas. Architectural compatibility between generative and discriminative models is ensured by using a single design pattern. The unified architecture of neural network models made it possible to apply the transfer learning method. The use of pretrained parameters in the variational autoencoder model makes it possible to compensate for the small size of the CSCDroneCL_v1 dataset. The generated synthetic training examples of panoramas were added to the VisDrone2022 dataset. Using the new dataset, a discriminative model was trained, resulting in an increase in anomaly classification accuracy by 21.2% for the selected class. [ABSTRACT FROM AUTHOR]

Cardiac arrhythmia identification and categorization are crucial for prompt treatment and better patient outcomes. Arrhythmia identification is the main focus of this study's temporal attention network (TAN)-based multiclass categorization of varied-length electrocardiogram (ECG) data. The suggested TAN is designed to handle variable-duration ECG signals, making it ideal for real-time monitoring. The TAN uses a dynamic snippet extraction approach to choose meaningful ECG segments to ensure the model captures essential properties despite the constraints of processing such heterogeneous data. Training and assessment use a large dataset of atrial fibrillation, ventricular, and supraventricular arrhythmias. The TAN outperforms current approaches in multiclass early arrhythmia classification and is very accurate. Concatenating EfficientNet with CNN layer helped overcome different data and variable-length signals. High accuracy: 98% of normal, 97.1% of atrial fibrillation (AF), 98% of other, and 98% of noisy using the proposed CEEC model. Early arrhythmia diagnosis has improved due to the TAN's ability to effectively identify varied-length ECG data and give interpretability. It enables quicker interventions, personalised treatment plans, and improved arrhythmia control, which can greatly benefit patient care. [ABSTRACT FROM AUTHOR]

Floods, caused by intense rainfall or typhoons, overwhelming urban drainage systems, pose significant threats to urban areas, leading to substantial economic losses and endangering human lives. This study proposes a methodology for flood assessment in urban areas using a multiclass classification approach with a Deep Neural Network (DNN) optimized through hyperparameter tuning with genetic algorithms (GAs) leveraging remote sensing data of a flood dataset for the Ibadan metropolis, Nigeria and Metro Manila, Philippines. The results show that the optimized DNN model significantly improves flood risk assessment accuracy (Ibadan-0.98) compared to datasets containing only location and precipitation data (Manila-0.38). By incorporating soil data into the model, as well as reducing the number of classes, it is able to predict flood risks more accurately, providing insights for proactive flood mitigation strategies and urban planning. [ABSTRACT FROM AUTHOR]

Fault classification in industrial processes is of paramount importance, as it allows the implementation of preventive and corrective measures before catastrophic failures occur, which can result in significant repair costs and production loss, for example. Therefore, the purpose of this study was to develop a classification model by merging the concepts of Decision Trees with Genetic Programming. To accomplish this, the proposed model randomly generates a set of decision trees using the adapted Tennessee Eastman dataset. The generation of these trees does not rely on classical construction logic; instead, they employ an approach where the structure and characteristics of the trees are randomly determined and adjusted throughout the evolutionary process. This approach enables a broader exploration of the search space and may lead to diverse solutions. The results obtained were moderate, largely due to the high number of target classes for classification (21 classes), resulting in the creation of complex trees. The average accuracy on the test data was 0.75, indicating the need to implement new alternatives and enhancements in the algorithm to improve the results.

Abstract Profiling social media users is an analytical approach to generate an extensive blueprint of user’s personal characteristics, which can be useful for a diverse range of applications, such as targeted marketing and personalized recommendations. Although social user profiling has gained substantial attention in recent years, effectively constructing a collaborative model that could describe long and short-term profiles is still challenging. In this paper, we will discuss the profiling problem from two perspectives; how to mathematically model and track user’s behavior over short and long periods and how to enhance the classification of user’s activities. Using mathematical equations, our model can define periods in which the user's interests abruptly changed. A dataset consisting of 30,000 tweets was built and manually annotated into 10 topic categories. Bi-LSTM and GRU models are applied to classify the user’s activities representing his interests, which then are utilized to create and model the dynamic profile. In addition, the effect of word embedding techniques and pre-trained classification models on the accuracy of the classification process is explored in this research.

Domain Generation Algorithms (DGAs) are algorithms present in most malware used by botnets and advanced persistent threats. These algorithms dynamically generate domain names to maintain and obfuscate communication between the infected device and the attacker's command and control server. Since DGAs are used by many threats, it is extremely important to classify a given DGA according to the threat it is related to. In addition, as new threats emerge daily, classifier models tend to become obsolete over time. Deep neural networks tend to lose their classification ability when retrained with a dataset that is significantly different from the initial one, a phenomenon known as catastrophic forgetting. This work presents a computational scheme composed of a deep learning model based on CNN and natural language processing and an incremental learning technique for class increment through transfer learning to classify 60 DGA families and include a new family to the classifier model, training the model incrementally using some examples from known families, avoiding catastrophic forgetting and maintaining metric levels. The proposed methodology achieved an average precision of 86.75%, an average recall of 83.06%, and an average F1 score of 83.78% with the full dataset, and suffered minimal losses when applying the class increment. [ABSTRACT FROM AUTHOR]

To tackle the common issue of imbalanced data classes in the fault diagnosis of avionics equipment, this study proposes a method that integrates the Synthetic Minority Oversampling Technique (SMOTE) with Boosting (SMOTEWB) and the Light Gradient Boosting Machine (LGBM). Initially, SMOTEWB is refined through a "successive one‐vs‐many balancing strategy," which effectively addresses multiclass sample balance issues. The enhanced data is then employed for pattern recognition and classification using LGBM. Additionally, this study utilizes the Tree‐structured Parzen Estimator (TPE) method and five‐fold cross‐validation to optimize the model's hyperparameters, thus improving diagnostic accuracy. Experimental validation using University of California Irvine (UCI) public datasets and real‐world avionics equipment fault data shows that the proposed SMOTEWB‐LGBM method outperforms other common methods in handling multiclass imbalanced datasets. This new approach not only enhances fault diagnosis efficiency but also offers a potent solution for similar multiclass imbalance challenges. [ABSTRACT FROM AUTHOR]

To solve the problems of low recognition accuracy and poor generalization performance of existing face anti-spoofing models in different application scenarios, this paper adopted the idea of disentangled representation learning and proposed a face anti-spoofing method based on disentangled representation learning. This method adopted U-Net architecture and ResNet18 as the encoder-decoder. In the first stage of training, it inputted real samples so that the encoder only learned information related to real samples. In the second stage, this paper built an adversarial learning network, inputted samples without labels, feature fusion of the pre-trained encoder output and the new encoder output, reconstructed the image by the decoder, and performed adversarial training with the original image in the discriminator to achieve feature decoupling. Compared with some classic face anti-spoofing methods, the model paper achieved better detection performance. The lowest detection error rate in several experiments on the OULU-NPU data set is only 0.8%, which is better than classic detection methods such as STDN. The face anti-spoofing method used staged training to enable the model to obtain a more discriminative feature representation than the end-to-end model in adversarial training. It adopted a multi-classification strategy in the deception feature map output stage to reduce the impact of different image noises on classification results, and experiments on public data sets verified the effectiveness of the algorithm. [ABSTRACT FROM AUTHOR]

Food demand is exponentially increasing due to the increase in population in every country; hence, increasing the yield is one of the focus areas for sustainable agricultural development. Predicting plant disease is one of the measures to increase crop yield and quality, thereby increasing the economy. The present work aims to develop a web-based application built with a deep learning model to detect plant leaf disease using a leaf image and alert farmers with messages. A comparative study was conducted on the data of the PlantVillage dataset for binary and multiclass classifications. Various deep convolutional neural network (CNN) models, such as MobileNet, DenseNet201, ResNet50, Inception V3, and visual geometry group (VGG) 16 and 19, have been compared with a proposed model. Various metrics include precision, recall, classification report, Confusion Matrix, and accuracy. MobileNet is influential among the selected models, with an accuracy of 97.35% and precision and recall of 0.973 each for multiclass classification. The proposed model achieved an accuracy of 99.39% with a loss of 0.0361, precision of 0.989, and a recall of 0.984 for binary classification compared with deep CNN models. A web-based application was created using the MobileNet model for the convenience of sending an email alert to the user regarding plant disease. The research results help improve a country's crop productivity and the overall economy through prompt and precise decision-making on crop diseases. [ABSTRACT FROM AUTHOR]

The ordered logit (OLogit) model is a regression model for an ordinal dependent variable. For a conventional time series OLogit model, both the dependent variable and the independent variables are required to be observed at the same frequency. However, this requirement is violated under the circumstance of mixed frequency data. To this end, we introduce the unrestricted MIDAS (U‐MIDAS) method into the OLogit model and develop a novel U‐MIDAS‐OLogit model, in which high‐frequency covariates are used to predict a low‐frequency outcome with ordinal categories. The U‐MIDAS‐OLogit model enlarges the application of OLogit and enables to produce timely forecast. To verify its effectiveness, we conduct extensive Monte Carlo simulations. The numerical results show that the U‐MIDAS‐OLogit model is superior to several typical OLogit models in terms of prediction performance. We then apply the U‐MIDAS‐OLogit model to predict credit ratings of listed companies in China and the US, respectively. The empirical results also confirm its promising in practical applications. [ABSTRACT FROM AUTHOR]

Various techniques have been employed to detect damage in civil engineering structures. Apart from the model-based approach, which demands the frequent updating of its corresponding finite element method (FEM)-built model, data-driven methods have gained prominence. Environmental and operational effects significantly affect damage detection due to the presence of damage-related trends in their analyses. Time-domain approaches such as autoregression and metrics such as the Mahalanobis squared distance have been utilized to mitigate these effects. In the realm of machine learning (ML) models, their effectiveness relies heavily on the type and quality of the extracted features, making this aspect a focal point of attention. The objective of this work is therefore to deploy and observe potential feature extraction approaches used as input in training fully data-driven damage detection machine learning models. The most damage-sensitive segment (MDSS) feature extraction technique, which potentially treats signals under multiple conditions, is also proposed and deployed. It identifies potential segments for each feature coefficient under a defined criterion. Therefore, 680 signals, each consisting of 8192 data points, are recorded using accelerometer sensors at the Los Alamos National Laboratory in the USA. The data are obtained from a three-story 3D building frame and are utilized in this research for a mainly data-driven damage detection task. Three approaches are implemented to replace four missing signals with the generated ones. In this paper, multiple fast Fourier and wavelet-transformed features are employed to evaluate their performance. Most importantly, a power spectral density (PSD)-based feature extraction approach that considers the maximum variability criterion to identify the most sensitive segments is developed and implemented. The performance of the MDSS selection technique, proposed in this work, surpasses that of all 18 trained neural networks (NN) and recurrent neural network (RNN) models, achieving more than 80% prediction accuracy on an unseen prediction dataset. It also significantly reduces the feature dimension. Furthermore, a sensitivity analysis is conducted on signal segmentation, overlapping, the treatment of a training dataset imbalance, and principal component analysis (PCA) implementation across various combinations of features. Binary and multiclass classification models are employed to primarily detect and additionally locate and identify the severity class of the damage. The collaborative approach of feature extraction and machine learning models effectively addresses the impact of environmental and operational effects (EOFs), suppressing their influences on the damage detection process. [ABSTRACT FROM AUTHOR]

Implementing predictive maintenance is paramount in guaranteeing the dependability and efficiency of intricate industrial systems. Classification approaches have been extensively utilized to detect and anticipate potential malfunctions in acknowledged systems. Numerous machine learning algorithms exhibit effectiveness, yet they encounter interpretability issues, posing a challenge in comprehending the fundamental factors contributing to their predictions. This study investigates the utilization of machine learning algorithms for classification in the domain of predictive maintenance. This research focuses on applying Random Forest, XGBoost, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) algorithms for Turbofan Engine classification into multiple categories. We use Explainable Artificial Intelligence (XAI) methodologies to mitigate the interpretability challenges commonly associated with black-box models. We aim to utilize XAI algorithms on the most effective models to elucidate the essential characteristics and decision-making procedures underlying the classification outcomes. The findings of our experiment indicate that the Random Forest algorithm based on ensemble learning and the XGBoost algorithm based on gradient boosting exhibit superior classification accuracy. Additionally, the utilization of Explainable Artificial Intelligence (XAI) algorithms amplifies the transparency and interpretability of said models, thereby facilitating a more profound comprehension of the determinants that impact the classification results. The results of this study make a valuable contribution to predictive maintenance by demonstrating the efficacy of machine learning algorithms in multiclass classification and emphasizing the significance of explainable artificial intelligence (XAI) in furnishing practical insights for decision-making. [ABSTRACT FROM AUTHOR]

Copyright of Dicle University Journal of Engineering / Dicle Üniversitesi Mühendislik Dergisi is the property of Dicle Universitesi and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

This study proposes a method for classifying economic activity descriptors to match Nomenclature of Economic Activities (NACE) codes, employing a blend of machine learning techniques and expert evaluation. By leveraging natural language processing (NLP) methods to vectorize activity descriptors and utilizing genetic algorithm (GA) optimization to fine-tune hyperparameters in multi-class classifiers like Naive Bayes, Decision Trees, Random Forests, and Multilayer Perceptrons, our aim is to boost the accuracy and reliability of an economic classification system. This system faces challenges due to the absence of precise target labels in the dataset. Hence, it is essential to initially check the accuracy of utilized methods based on expert evaluations using a small dataset before generalizing to a larger one. [ABSTRACT FROM AUTHOR]

Various techniques have been employed to detect damage in civil engineering structures. Apart from the model-based approach, which demands the frequent updating of its corresponding finite element method (FEM)-built model, data-driven methods have gained prominence. Environmental and operational effects significantly affect damage detection due to the presence of damage-related trends in their analyses. Time-domain approaches such as autoregression and metrics such as the Mahalanobis squared distance have been utilized to mitigate these effects. In the realm of machine learning (ML) models, their effectiveness relies heavily on the type and quality of the extracted features, making this aspect a focal point of attention. The objective of this work is therefore to deploy and observe potential feature extraction approaches used as input in training fully data-driven damage detection machine learning models. The most damage-sensitive segment (MDSS) feature extraction technique, which potentially treats signals under multiple conditions, is also proposed and deployed. It identifies potential segments for each feature coefficient under a defined criterion. Therefore, 680 signals, each consisting of 8192 data points, are recorded using accelerometer sensors at the Los Alamos National Laboratory in the USA. The data are obtained from a three-story 3D building frame and are utilized in this research for a mainly data-driven damage detection task. Three approaches are implemented to replace four missing signals with the generated ones. In this paper, multiple fast Fourier and wavelet-transformed features are employed to evaluate their performance. Most importantly, a power spectral density (PSD)-based feature extraction approach that considers the maximum variability criterion to identify the most sensitive segments is developed and implemented. The performance of the MDSS selection technique, proposed in this work, surpasses that of all 18 trained neural networks (NN) and recurrent neural network (RNN) models, achieving more than 80% prediction accuracy on an unseen prediction dataset. It also significantly reduces the feature dimension. Furthermore, a sensitivity analysis is conducted on signal segmentation, overlapping, the treatment of a training dataset imbalance, and principal component analysis (PCA) implementation across various combinations of features. Binary and multiclass classification models are employed to primarily detect and additionally locate and identify the severity class of the damage. The collaborative approach of feature extraction and machine learning models effectively addresses the impact of environmental and operational effects (EOFs), suppressing their influences on the damage detection process.

Social media platforms have gained immense popularity in recent years and are used for various activities such as marketing, news-sharing, and celebrating achievements. However, they are also notorious for spreading hateful and discriminatory content, which can cause harm to individuals and communities. Therefore, it is crucial to detect and remove such content from social media platforms as soon as possible. Although research related to the detection of hate speech and inflammatory content is increasing, studies focused on code-mixed Indian languages are limited. Hence, in this work, we have conducted a comprehensive study, where we have compared the effectiveness of various neural networks, and transformer-based techniques for the detection of hate and objectionable language in social media tweets in Hinglish, Tamil written in English, and Malayalam written in English, to propose the best-performing ensemble model, named as DConvBLSTM-MuRIL. To carry out our experiments, we have created our datasets for the three languages under study and compared the results with already existing datasets. Our proposed weighted ensemble framework outperformed the existing models, achieving better-weighted F1-scores and better accuracy for all the three languages under consideration. [ABSTRACT FROM AUTHOR]

Drones are widespread and actively employed in a variety of applications due to their low cost and quick mobility and enabling new forms of action surveillance. However, owing to various challenges- limited no. of aerial view samples, aerial footage suffers with camera motion, illumination changes, small actor size, occlusion, complex backgrounds, and varying view angles, human action recognition in aerial videos even more challenging. Maneuvering the same, we propose Aerial Polarized-Transformer Network (AP-TransNet) to recognize human actions in aerial view using both spatial and temporal details of the video feed. In this paper, we present the Polarized Encoding Block that performs ( i) Selection with Rejection to select the significant features and reject least informative features similar to Light photometry phenomena and ( ii) boosting operation increases the dynamic range of encodings using non-linear softmax normalization at the bottleneck tensors in both channel and spatial sequential branches. The performance of the proposed AP-TransNet is evaluated by conducting extensive experiments on three publicly available benchmark datasets: drone action dataset, UCF-ARG Dataset and Multi-View Outdoor Dataset (MOD20) supporting with ablation study. The proposed work outperformed the state-of-the-arts. [ABSTRACT FROM AUTHOR]

Personalized medicine enables precise tumor treatment for a patient's molecular genetic profile. To devise optimal targeted treatment plans for patients in a molecular tumor board, physicians must consider alterations on gene- and proteins levels but also cancer cell phenotypes. Machine learning can uncover buried patterns, extract pivotal information, and unveil corresponding insights from available data. Publicly available datasets provide the amounts of data necessary. This work outlines the efficacy of various machine learning algorithms which could eventually serve as clinical decision support in a precision oncology setting. Leveraging algorithms including Random Forest, Decision tree, XGBoost, Logistic regression, Gaussian Naive Bayes, k nearest neighbor, and AdaBoost, we conducted two experiments for the breast invasive carcinoma dataset. Incorporated data includes patient-, molecular- and treatment data. The aim of the investigation was to predict medication treatment or type of treatment based on genetic profile. After preprocessing and application of ML algorithms, the first results were promising. Multiple factors challenge application in clinical care settings without carefully considering the limitations. [ABSTRACT FROM AUTHOR]

The selection of group features is a critical aspect in reducing model complexity by choosing the most essential group features, while eliminating the less significant ones. The existing group feature selection methods select a set of important group features, without providing the relative importance of all group features. Moreover, few methods consider the relative importance of group features in the selection process. This study introduces a permutation-based group feature selection approach specifically designed for high-dimensional multiclass datasets. Initially, the least absolute shrinkage and selection operator (lasso) method was applied to eliminate irrelevant individual features within each group feature. Subsequently, the relative importance of the group features was computed using a random-forest-based permutation method. Accordingly, the process selected the highly significant group features. The performance of the proposed method was evaluated using machine learning algorithms and compared with the performance of other approaches, such as group lasso. We used real-world, high-dimensional, multiclass microarray datasets to demonstrate its effectiveness. The results highlighted the capability of the proposed method, which not only selected significant group features but also provided the relative importance and ranking of all group features. Furthermore, the proposed method outperformed the existing method in terms of accuracy and F1 score. [ABSTRACT FROM AUTHOR]

Fourier transform infrared spectroscopy (FT-IR) is a widely used spectroscopic method for routine analysis of substances and compounds. Spectral interpretation of spectra is a labor-intensive process that provides important information about functional groups or bonds present in compounds and complex substances. In this paper, based on deep learning methods of convolutional neural networks, models were developed to determine the presence of 17 classes of functional groups or 72 classes of coupling oscillations in the FT-IR spectra. Using web scanning, the spectra of 14 361 FT-IR spectra of organic molecules were obtained. Several different variants of model architectures with different sizes of feature maps have been tested. Based on the Shapley additive explanations (SHAP) and gradient-weighted class activation mapping (GradCAM) methods, visualization tools have been developed for visualizing and highlighting the areas of absorption bands manifestation for corresponding functional groups or bonds in the spectrum. To determine 17 and 72 classes, the F1-weighted metric, which is the harmonic mean of the class' precision and class' recall weighted by class' fraction, reached 93 and 88%, respectively, when using data on the position of absorption maxima in the spectrum as an additional source layer. The resulting model can be used to facilitate the routine analysis of spectra for all areas such as organic chemistry, materials science, and biology, as well as to facilitate the preparation of the obtained experimental data for publication. [ABSTRACT FROM AUTHOR]

Handwritten documents possess immense significance in domains such as law, history, and administration. However, they are vulnerable to forgery, which can undermine their credibility and reliability. This paper aims to establish a dependable technique for identifying altered text in handwritten document images, even in scenarios with high levels of noise and blur. Our study investigates 10 distinct categories of handwritten text that have been altered through various forgery operations. The suggested approach employs the deep neural architectures VGG16 and Resnet50 as feature extractors. The architecture comprises three parts: Feature extraction using individual models, a feature fusion layer, and a classification layer. Initially, we optimize the training process and feature extraction using VGG16 and ResNet50. The feature vectors obtained from both models are then fused together in the feature fusion layer and input into the classification layer for the classification task. Experiments are conducted on a custom-created dataset as well as benchmark datasets including ICPR FDC, IMEI Forged Number, and Kundu to demonstrate that the proposed method is superior to existing approaches. [ABSTRACT FROM AUTHOR]

Gastrointestinal diseases cause around two million deaths globally. Wireless capsule endoscopy is a recent advancement in medical imaging, but manual diagnosis is challenging due to the large number of images generated. This has led to research into computer-assisted methodologies for diagnosing these images. Endoscopy produces thousands of frames for each patient, making manual examination difficult, laborious, and error-prone. An automated approach is essential to speed up the diagnosis process, reduce costs, and potentially save lives. This study proposes transfer learning-based efficient deep learning methods for detecting gastrointestinal disorders from multiple modalities, aiming to detect gastrointestinal diseases with superior accuracy and reduce the efforts and costs of medical experts. The Kvasir eight-class dataset was used for the experiment, where endoscopic images were preprocessed and enriched with augmentation techniques. An EfficientNet model was optimized via transfer learning and fine tuning, and the model was compared to the most widely used pre-trained deep learning models. The model's efficacy was tested on another independent endoscopic dataset to prove its robustness and reliability. [ABSTRACT FROM AUTHOR]

The detection of weed species in rangeland environments is a challenging task due to various factors such as dense, variable species vegetation, ocular occlusion, and a wide variety of plant morphology. Most research in weed detection, however, focuses on croplands. This research addresses the need for accurate rangeland weed detection models by leveraging convolutional neural network (CNN) models enhanced with transfer learning applied to the DeepWeeds data set taken in situ in regional North Eastern Australia. It investigates the effectiveness of transfer learning across seven popular models, utilizing data augmentation and fine-tuning. The performance of these models was evaluated using accuracy metrics and compared against each other. The results demonstrated that transfer learning, coupled with fine tuning, could be a viable solution for generating efficient weed plant detection models with lower demands on computational resources and smaller datasets, despite the challenging conditions of rangeland environments. EfficientNetV2B1 had the highest classification accuracy of 94.2 ​%, and lowest training times. Moreover, high levels of accuracy were also achieved using InceptionV3, VGG16, and Densenet121, albeit with a training time penalty. This research provides insights into the performance of CNN models in challenging rangeland environments, demonstrates the potential of using transfer learning to enhance weed detection models, and underscores the significance of model selection in agricultural applications of CNNs.

The diagnosis and classification of autism spectrum disorder (ASD) presents anatomical difficulty owing to the existence of a wide range of symptoms that may be organized into many categories. The present research investigates the efficacy of machine learning methods for facilitating the recognition of individuals who have been diagnosed with ASD. The primary aim of this study has been to assess the effectiveness of multiple algorithms based on machine learning in identifying intricate patterns seen in datasets related to ASD, which includes a wide range of diagnostic categories. The results indicate that the Logistic Regression approach demonstrated great levels of accuracy, with rates of 94.3 % for children and 99 % for adolescents in the binary classification system. Similarly, it has been reported that the Support Vector Machine (SVM) had superior performance compared to all other systems in the binary classification test focused on adults exclusively, with an accuracy rate of 98.5 %. Moreover, a supplementary series of experiments conducted on the combined dataset of children, adolescents, and adults has resulted in the observation that Logistic Regression and SVM exhibited notable accuracy rates of 97.2 % for binary classification and 99.55 % for multiclass classification, encompassing individuals from diverse age groups. The results provide evidence in favor of that progress has been achieved in the diagnosis and treatment of ASD as a result of the capacity to detect and categorize the disorder at an earlier developmental phase.

Liver disease includes a range of health conditions resulting from various factors that impact the normal functioning of the liver over an extended period. Effective treatment of Hepatitis C, a type of liver disease, can be significantly enhanced by accurately and timely predicting risk factors and severity, covering different stages like fibrosis and cirrhosis. This research utilizes the Hepatitis C dataset from the UCI repository to present a comprehensive framework for the prediction of liver disease across various stages. We have proposed an adaptive data preprocessing technique designed to enhance the efficacy of our foundational ML models. This method incorporates class specific mean imputation, outlier rejection, log normalization, feature selection, feature scaling and data balancing. The proposed preprocessing technique underwent rigorous analysis. Furthermore, we also proposed a set of ensemble models by combining basic ML classifiers to further improve the accuracy of liver disease predictions. The best model underwent rigorous parameter optimization and exhibited remarkable training and testing accuracy, reaching up to 99.87% and 99.80% respectively, outperforming earlier research on the dataset. A user-friendly interface was developed to enhance user interaction and practical use, enabling medical professionals to effortlessly enter patient information and promptly receive assessments regarding the risk factors associated with liver disorders.

Abstract Background The mRNA subcellular localization bears substantial impact in the regulation of gene expression, cellular migration, and adaptation. However, the methods employed for experimental determination of this localization are arduous, time-intensive, and come with a high cost. Methods In this research article, we tackle the essential challenge of predicting the subcellular location of messenger RNAs (mRNAs) through Unified mRNA Subcellular Localization Predictor (UMSLP), a machine learning (ML) based approach. We embrace an in silico strategy that incorporate four distinct feature sets: kmer, pseudo k-tuple nucleotide composition, nucleotide physicochemical attributes, and the 3D sequence depiction achieved via Z-curve transformation for predicting subcellular localization in benchmark dataset across five distinct subcellular locales, encompassing nucleus, cytoplasm, extracellular region (ExR), mitochondria, and endoplasmic reticulum (ER). Results The proposed ML model UMSLP attains cutting-edge outcomes in predicting mRNA subcellular localization. On independent testing dataset, UMSLP ahcieved over 87% precision, 94% specificity, and 94% accuracy. Compared to other existing tools, UMSLP outperformed mRNALocator, mRNALoc, and SubLocEP by 11%, 21%, and 32%, respectively on average prediction accuracy for all five locales. SHapley Additive exPlanations analysis highlights the dominance of k-mer features in predicting cytoplasm, nucleus, ER, and ExR localizations, while Z-curve based features play pivotal roles in mitochondria subcellular localization detection. Availability We have shared datasets, code, Docker API for users in GitHub at: https://github.com/smusleh/UMSLP .

Abstract Computational identification and classification of clinical disorders gather major importance due to the effective improvement of machine learning methodologies. Cancer identification and classification are essential clinical areas to address, where accurate classification for multiple types of cancer is still in a progressive stage. In this article, we propose a multiclass cancer classification model that categorizes the five different types of cancers using gene expression data. To perform efficient analysis of the available clinical data, we propose feature selection and classification methods. We propose a genetic clustering algorithm (GCA) for optimal feature selection from the RNA-gene expression data, consisting of 801 samples belonging to the five major classes of cancer. The proposed feature selection method reduces the 1621 gene expressions into a cluster of 21 features. The optimum feature set acts as input data to the proposed divergent random forest. Based on the features computed, the proposed classifier categorizes the data samples into 5 different classes of cancers, including breast cancer, colon cancer, kidney cancer, lung cancer, and prostate cancer. The proposed divergent random forest provided performance improvisation in terms of accuracy with 95.21%, specificity with 93%, and sensitivity with 94.29% which outperformed all the other existing multiclass classification algorithms.

Tool condition monitoring (TCM) is crucial to ensure good quality products and avoid downtime. Machine learning has proven to be vital for TCM. However, existing works are predominately based on supervised learning, which hinders their applicability in real-world manufacturing settings, where data labeling is cumbersome and costly with in-service machines. Additionally, the existing unsupervised solutions mostly handle binary decision-based TCM which is unable to fully reflect the dynamics of tool wear progression. To address these issues, we propose different unsupervised and semisupervised five-class tool wear recognition frameworks to handle fully unlabeled and partially labeled data, respectively. The underlying methods include Laplacian score, sparse autoencoder (SAE), stacked SAE (SSAE), self-organizing map, Softmax, support vector machine, and random forest. For the semisupervised frameworks, we considered designs where labeled data influence only feature learning, classifier building, or both. We also investigated different training configurations of SSAE regarding the supervision level. We applied the frameworks on two run-to-failure datasets of milling tools, recorded using a microphone and an accelerometer. Single sensor and multisensor data under different percentages of labeled training data were considered in the evaluation. The results showed which of the frameworks led to the best predictive performance under which data settings, and highlighted the significance of sensor fusion and discriminative feature representations in combating the unavailability and scarcity of labels, among other findings. The highest macro-F1 achieved for the two datasets with fully unlabeled data reached 87.52% and 75.80%, respectively, and over 90% when only 25% of the training observations were labeled.