Parkinson's disease (PD) is a progressive neurodegenerative disorder, and the diagnostic procedures are very crucial in enhancing patient outcomes when performed on time and accurately. Machine learning procedures have shown promising results in identifying PD; however, the small sample sizes and strong imbalance of the classes, with substantially more patient samples than healthy ones, are often observed in voice features-based biomedical data from Parkinson’s patients, which poses significant limitations to the model generalization and stability. The current study was motivated by the need to conduct an objective evaluation of the influence of data preprocessing and model-level approaches on performance under these constraints. In this regard, three scenario situations were developed. Recursive Feature Elimination was used first to reduce 18 salient features in the first two scenarios, and then resampling methods at the data level were applied: Instance Hardness Threshold undersampling and a hybrid oversampling regime comprising K-means Synthetic Minority Oversampling Technique (SMOTE), Borderline-SMOTE, and SMOTE-Tomek. In spite of the fact that these approaches moderated the imbalance between classes, they brought up side effects like loss of information and distortion of decision limits that were especially acute considering the small size and sensitivity of the PD data. To address these constraints, the third scenario adopted a feature selection method using Fisher score, which was found to be beneficial in reducing the feature redundancy, together with algorithm-level imbalance reduction using highly fine-tuned CatBoost and Support Vector Machine models, which were trained and evaluated to classify PD cases. This plan took advantage of the discriminative ability of the fine feature set and maintained data integrity. The results indicate that CatBoost in the third case achieved the best performance metrics (accuracy = 97%, area under the curve = 0.96, F1 = 0.98), and this in turn supports the fact that the combination of feature-level refinement and algorithmic adaptation is a comparatively stronger performance under benchmark evaluation conditions for diagnosis of PD. The study can be identified by its comprehensive design since it analyzes all of the scenarios systematically with equal experimental conditions and various train-test splits.

In this study, an efficient approach utilizing the Minkowski distance metric is introduced for the identification of arrhythmias through the examination of electrocardiogram (ECG) signals. Current machine learning classifiers, such as Naive Bayes (NB), Decision Trees (DT), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF), often face challenges in handling high-dimensional features and capturing complex relationships within signals, which are critical for accurate arrhythmia detection. The proposed method employs the Minkowski metric as a classifier to measure the distance between ECG signals, leveraging its sensitivity to capture relationships and matching patterns within the data. This approach enables the accurate determination of the state of ECG signals, facilitating precise arrhythmia detection. Particle Swarm Optimization (PSO) is employed to select and optimize features, improving the proposed classifier’s capability to detect subtle patterns within ECG signals. The study utilizes the widely recognized MIT-BIH Arrhythmia dataset for evaluation. Performance metrics, including accuracy, precision, recall, and F1-score, are employed to assess the effectiveness of the proposed models. Experimental results demonstrate that the proposed approach outperforms traditional classifiers such as NB, SVM, KNN, RF, and DT, achieving accuracy rates of up to 91.25% and 93.75% without and with PSO, respectively. These findings highlight the potential of the Minkowski-based technique in improving arrhythmia detection and addressing the limitations of traditional ML methods. With its ability to support early arrhythmia diagnosis, the proposed approach presents a valuable and efficient solution for medical practitioners, contributing to better clinical outcomes.

This paper aims to introduce a novel clustering method for electroencephalogram (EEG) based on Ruzicka mathematical similarity and incorporates Particle Swarm Optimization (PSO) to enhance feature selection. Medical datasets often contain both convergent and divergent features, making feature selection a crucial step for accurate disease diagnosis and public health applications. The proposed Ruzicka-based clustering method groups EEG records into non-overlapping subgroups according to a defined similarity metric. Cluster centers are determined using a polynomial-based calculation, after which EEG records are assigned to clusters based on the Ruzicka similarity measure. After clustering the EEG records into highly coherent groups, PSO algorithm is employed to identify the most effective subset of features. This process enhances classification accuracy and contributes to more reliable diagnostic outcomes by combining clustering with feature selection. The selected features are then evaluated using multiple classifiers, including Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN), and Naive Bayes (NB). Accuracy, recall, f1-score and precision measures are conducted to evaluate the model’s performance. Experimental validation is carried out on the Bonn University EEG dataset. With both RF and NB classifiers, the proposed model has achieved up to 100% accuracy compared to other models. The proposed method can be implemented in medical organizations as a decision-support system to assist healthcare professionals in analyzing EEG patterns. Its integration can enhance the accuracy and efficiency of disease diagnosis, leading to improved patient care.

In an industrial environment, sound is generally an important source of data that results from the movement and vibration of machines during operation. Each machine operation condition often generates a specific sound, for example, a familiar normal working sound, an unfamiliar sound, a breaking sound, or even the absence of sound, which may indicate a stop. This paper proposes a supervised convolutional neural network (CNN) model with early stopping and learning rate (LR) scheduling to achieve a more accurate and cost-effective trained deep learning (DL) model for machine sound classification. The proposed model is trained and evaluated using the MIMII dataset, which contains unbalanced real-world data and signal-to-noise ratios (SNRs). To ensure a fair (i.e., representative) data splitting strategy and robust coverage across all hierarchical structures of the MIMII dataset, this study adopts a hierarchical multi-stage stratified splitting protocol incorporating pre-split shuffling to prevent order bias. The synthetic minority oversampling technique (SMOTE) is used to balance the training set after feature extraction. Compared with state-of-the-art works implemented on the same dataset, the proposed model achieves optimal performance in terms of accuracy, area under the curve (AUC) , F1-score, precision, recall, matthews correlation coefficient (MCC), and Kappa without noise processing methods at an SNR of 6 dB, and maintains robust, competitive performance under severe noise conditions (0 dB and −6 dB).

This paper introduces a method for fake accounts detection on social media through machine learning using several classification algorithms and sophisticated feature extraction. Preprocessing steps included dataset integration, feature normalization, entropy-based feature extraction, behavioral ratio computation, and class imbalance handling using SMOTE and SMOTEENN. Ensemble learning models, including LightGBM, CatBoost, XGBoost, and Logistic Regression, were trained on the training dataset and subsequently evaluated on the unseen testing dataset to assess their classification performance. Models were evaluated using accuracy, precision, recall, F1 score, and confusion matrices. SHAP explanation displayed the feature contributions. Experimental results demonstrated the effectiveness of the proposed approach, achieving the highest predictive accuracy of 99.71

This paper explores the combined effect of diverse datasets and advanced feature extraction techniques on the performance of email spam detection using unsupervised clustering algorithms Fuzzy C-Means and K-Means. Leveraging the Enron Email Dataset and the SMS Spam Collection Dataset, we extract both traditional features (TF-IDF) and novel attributes (punctuation count, capital letter usage, and body length) that reflect spam characteristics. Clustering is applied to group emails based on these features, and the generated cluster labels are subsequently used to train supervised classifiers, including K-Nearest Neighbours, Naive Bayes, Support Vector Machines, and XGBoost. Comparative analysis reveals that clustering significantly enhances classification accuracy by uncovering latent structures within high-dimensional data. Moreover, our results highlight the strengths and limitations of Fuzzy C-Means versus K-Means in this context. The study demonstrates that integrating unsupervised clustering with rich feature engineering contributes to more robust and efficient spam detection, paving the way for future hybrid spam filtering models.

Standard classifiers struggle with high-dimensional datasets due to increased computational complexity, difficulty in visualization and interpretation, and challenges in handling redundant or irrelevant features. This paper proposes a novel feature selection method based on the Mahalanobis distance for Parkinson's disease (PD) classification. The proposed feature selection identifies relevant features by measuring their distance from the dataset's mean vector, considering the covariance structure. Features with larger Mahalanobis distances are deemed more relevant as they exhibit greater discriminative power relative to the dataset's distribution, aiding in effective feature subset selection. Significant improvements in classification performance were observed across all models. On the "Parkinson Disease Classification Dataset", the feature set was reduced from 22 to 11 features, resulting in accuracy improvements ranging from 10.17% to 20.34%, with the K-Nearest Neighbors (KNN) classifier achieving the highest accuracy of 98.31%. Similarly, on the "Parkinson Dataset with Replicated Acoustic Features", the feature set was reduced from 45 to 18 features, achieving accuracy improvements ranging from 1.38% to 13.88%, with the Random Forest (RF) classifier achieving the best accuracy of 95.83%. By identifying convergence features and eliminating divergence features, the proposed method effectively reduces dimensionality while maintaining or improving classifier performance. Additionally, the proposed feature selection method significantly reduces execution time, making it highly suitable for real-time applications in medical diagnostics, where timely and accurate disease identification is critical for improving patient outcomes.

The rapid and increasing growth in the volume and number of cyber threats from malware is not a real danger; the real threat lies in the obfuscation of these cyberattacks, as they constantly change their behavior, making detection more difficult. Numerous researchers and developers have devoted considerable attention to this topic; however, the research field has not yet been fully saturated with high-quality studies that address these problems. For this reason, this paper presents a novel multi-objective Markov-enhanced adaptive whale optimization (MOMEAWO) cybersecurity model to improve the classification of binary and multi-class malware threats through the proposed MOMEAWO approach. The proposed MOMEAWO cybersecurity model aims to provide an innovative solution for analyzing, detecting, and classifying the behavior of obfuscated malware within their respective families. The proposed model includes three classification types: binary classification and multi-class classification (e.g., four families and 16 malware families). To evaluate the performance of this model, we used a recently published dataset called the Canadian Institute for Cybersecurity Malware Memory Analysis (CIC-MalMem-2022) that contains balanced data. The results show near-perfect accuracy in binary classification and high accuracy in multi-class classification compared with related work using the same dataset.

This study introduces a novel feature selection method based on Hellinger distance and particle swarm optimization (PSO) for reducing the dimensionality of features in electroencephalogram (EEG) signals and improving epileptic seizure detection accuracy. In the first phase, the Hellinger distance is used as a filter to remove redundant and irrelevant features by calculating the similarity between blocks within the feature, thus reducing the search space for the subsequent second phase. In the second phase, PSO searches the reduced feature space to select the best subset. Recognizing that both classification accuracy and dimensionality play crucial roles in the performance of feature subsets, PSO searches various sets of features (ranging from 410 to 2867 in EEG signals) derived from the first stage using Hellinger distance, rather than searching through the full set of 4047 features, to select the optimal subset. The proposed Hellinger-PSO approach demonstrates significant improvements in classification accuracy across multiple models. Specifically, Logistic Regression (LR) improved from 91% to 95% (4% improvement), Decision Tree (DT) from 95% to 97% (2% improvement), Naive Bayes (NB) from 94% to 99% (5% improvement), and Random Forest (RF) from 96% to 98% (2% improvement) on the Bonn dataset. Additionally, the method reduces dimensionality while maintaining high classification performance. The results validate the efficacy of the Hellinger-PSO technique, which enhances both the accuracy and efficiency of epileptic seizure detection. This approach has the potential to improve diagnostic accuracy in medical settings, aiding in better patient care and more effective clinical decision-making.

In this paper, a novel classifier based on Robert’s similarity measure is introduced for emotion detection using electroencephalogram (EEG) signals. Traditional machine learning classifiers machine learning such as k-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Logistic Regression (LR), and Random Forest (RF), often struggle to accurately capture both linear and nonlinear patterns in EEG signals and face limitations in handling high-dimensional datasets. The proposed classifier addresses these challenges by segmenting EEG signals into block sizes categorized as small (1 to 10 samples), medium (20 to 100 samples), and large (200 to 1,000 samples), demonstrating particularly strong performance with medium and large block sizes to capture essential features. Integration of Particle Swarm Optimization (PSO) for feature selection, with Robert’s similarity as the fitness function, effectively refines the feature set, boosting classification accuracy and computational efficiency. Evaluation on an EEG brainwave dataset demonstrated that the method achieved an accuracy of 98.75% with feature selection, compared to 94.04% without it in emotional state detection. The results demonstrate that the proposed classifier is a valuable tool for diverse fields, including healthcare by detecting patient stress, education by assessing student engagement, customer service by monitoring satisfaction, and smart environments by enabling adaptive responses. Furthermore, the classifier has potential for broader industrial applications, such as improving workplace productivity by monitoring employee stress and enhancing safety in autonomous vehicle systems, making it a versatile solution for emotionally-aware systems across multiple domains.

Cloud computing environments are increasingly popular due to their flexibility and scalability, but they also present significant security challenges, particularly in the form of malware attacks. These malicious attacks exploit weaknesses within cloud infrastructures, which can result in serious repercussions like data breaches, unauthorized system access, and identity theft. In this paper, we introduce an innovative malware detection classifier specifically designed to overcome the shortcomings of conventional machine learning algorithms, such as K-Nearest Neighbor (KNN) and Support Vector Machine (SVM), in the unique context of cloud environments. Our proposed method relies on Log-spectral distance as a fundamental metric, which enables a more precise and effective approach to detecting malware. Through rigorous and extensive experimentation, our findings demonstrate that this novel classifier achieves an outstanding accuracy rate of 97

Phishing attacks continue to pose significant security risks, necessitating effective detection methods to safeguard users. This paper presents HawkPhish-DNN cybersecurity model, a novel phishing detection framework that integrates Harris Hawk Optimization (HHO) with a Deep Neural Network (DNN). In the preprocessing phase, redundant URLs and domain features are removed, while URL length and entropy are extracted to form an efficient feature set. The detection model employs advanced neural layers, including Sigmoid and ReLU, to enhance learning and classification. By leveraging multi-objective HHO, HawkPhish-DNN optimizes accuracy and mitigates false positives through a time-varying penalty function, Pareto dominance, and crowding distance strategies. Empirical findings demonstrate that HawkPhish-DNN cybersecurity model achieves an accuracy of up to 99.6% and a false positive rate as low as 0.2% on benchmark datasets. Additionally, it maintains low computational overhead, ensuring practicality for real-time deployment. These results highlight the potential of HawkPhish-DNN cybersecurity model in providing a robust, user-friendly defense against phishing threats without inflating false alarms.

Environmental pollution and waterborne diseases underscore the need for accurate water quality prediction to ensure public health. Therefore, Intelligent algorithms play a crucial role in estimating contaminant levels. However, the presence of outliers in the data can hurt the predictions. Therefore, dealing with these outliers is essential. In addition, regression algorithms often have problems with predicting data with overlapping features. This paper proposes an Adaptive Clustering Regression (ACR) model to enhance water quality index prediction. The proposed model has three stages: feature selection, data segmentation, and prediction. In the feature selection stage, the proposed model selects features based on their high correlation with the target variable (water quality index). The data is dynamically subdivided in the second phase based on density. The model trains a specific regression algorithm on each cluster. Evaluations of datasets from 37 monitoring stations demonstrate that ACR outperforms existing regression models, achieving over 5% improvement in predictive accuracy. The proposed model achieved good results and an improvement of over 5% compared to other regression algorithms.

Distributed Denial of Service (DDoS) attacks continue to overwhelm networked systems, demanding detectors that are accurate, low-false-alarm, transferable, and deployable. We propose OSES-DL, an Optimization-guided Statistical Ensemble Synergistic Deep Learning framework that advances all four fronts. The method introduces: (i) an Optimization-driven Feature Evolution Layer (OFEL) that co-trains feature sparsity with accuracy, stability, and entropy preservation; (ii) a Statistical Deep Synergy Module (SDSM) that injects Mahalanobis anomaly priors directly into BiLSTM hidden states, yielding anomaly-aware representations; (iii) Ensemble Knowledge Distillation with class-conditional temperature and feature–logit coupling (EKD-CCT) for calibrated, lightweight deployment; and (iv) a Cross-Domain Generalization Regularizer (CDGR) that combines prior-weighted MMD and CORAL for layer wise domain alignment. On CICDDoS2019, OSES-DL attains 99.45% accuracy, F1 0.994, AUC 0.998, and FAR 0.62%, with ECE 0.9%. Trained on CICDDoS2019 and tested on UNSW-NB15 and CAIDA, it improves F1 by +1.0% and reduces FAR by 0.5%–0.6% over the strongest baseline, while maintaining near-BiLSTM latency. Leave-one-attack-type-out tests confirm robustness to unseen vectors. Ablations attribute FAR reduction to SDSM, calibration to OFEL/EKD, and transferability to CDGR. OSES-DL delivers a principled, operationally grounded detector that is both state-of-the-art and deployment-ready.

This paper presents a novel classifier based on the Clark distance for early detection of Parkinson's disease (PD) using voice data. The nonlinear nature of human voice signals and their inherent fluctuations pose significant challenges for traditional machine learning classifiers such as Random Forest (RF), Support Vector Machines (SVM), Decision Trees (DT), K-Nearest Neighbors (KNN), and Logistic Regression (LR), which struggle to capture meaningful relationships within the voice data for accurate classification. The proposed classifier addresses these limitations by segmenting the data features of each voice sample into smaller blocks (ranging in size from 2 to 10 features per block), to better capture relationships and minimize fluctuations that negatively impact classification accuracy. To further enhance the performance of the classifier, Grey Wolf Optimization (GWO) was integrated to select high-harmony (convergent) features while removing irrelevant and redundant (divergent) features. The proposed approach achieved an accuracy of 94.6% without GWO and an impressive 98.305% accuracy with GWO, using only 12 selected features. Additionally, the classifier demonstrates efficient execution time, making it well-suited for real-time applications in medical organizations. The combination of the Clark distance classifier with GWO not only improves classification accuracy but also ensures computational efficiency, enabling its deployment in diverse settings such as hospitals and home monitoring systems for early detection and continuous monitoring of PD patients. This study highlights the potential of the proposed approach as a reliable, cost-effective, and scalable solution for voice-based PD detection.

This work describes a novel feature selection approach for detecting epileptic seizures in an EEG dataset that is based on a Hilbert similarity measurement of a convex set. Because the medical dataset has a high dimensionality, feature selection is vital for identifying diseases early and protecting human health. Moreover, high-dimensional data affect the prediction accuracy of machine learning algorithms and increase the system's complexity, making the results inefficient. Therefore, 1: This research presents a feature selection model mainly based on 2: Hilbert mathematical similarity measurement for computing similarity and related features based on high harmony inside the same feature. Using the proposed model, 3: The similarity between signals is computed and optimal, and 4: High similarity features are selected, while 5: Ineffective features are removed. Using the EEG data from Bonn University, we assessed the performance of the system. In addition, we assessed the proposed model based on recall, precision, and accuracy, which was compared to other earlier methods. Research findings confirmed that it was effective at extracting optimum features from EEG data. The proposed model obtained 100

This study presents an efficient feature selection method based on the Gower distance to enhance the accuracy and efficiency of standard classifiers on high-dimensional medical datasets. High-dimensional data poses significant challenges for traditional classifiers due to feature redundancy or being irrelevant. The proposed method addresses these challenges by partitioning the dataset into blocks, calculating the Gower distance within each block, and selecting features based on their average similarity. Technically, the Gower distance normalizes the absolute difference between numerical features, ensuring that each feature contributes equally to the distance calculation. This normalization prevents features with larger scales from overshadowing those with smaller scales. This process facilitates the identification of features that exhibit high harmony and are the most relevant for classification. The proposed feature selection strategy significantly reduces dimensionality, retains the most relevant features, and improves model performance. Experimental results show that the accuracy for the classifiers including k-nearest neighbors (KNN), naive Bayes (NB), decision tree (DT), random forest (RF), support vector machine (SVM), and logistic regression (LR) was increased by 4.38%–7.02%. Besides, the reduction in the feature set size contributes to a considerable decrease in computational complexity and thus faster diagnosis speed. The execution time was averagely reduced by 77.82% for all samples and 76.45% for one sample. These results demonstrate that the proposed feature selection method shows enhanced performance on both prediction accuracy and diagnostic speed, making it a promising tool for real-time clinical decision-making and improving patient care outcomes.

This paper proposes a novel clustering approach based on Gower distance and a new centroid cluster selection technique. Traditional clustering methods often struggle to improve the performance of standard classifiers when dealing with high-dimensional and heterogeneous data, leading to suboptimal outcomes. Moreover, methods like K-means are highly dependent on the initial selection of centers, which must be predefined, and they fail to effectively identify complex manifold clusters. The introduced approach overcomes these problems by segmenting records into blocks having 2–10 samples to understand their distribution and, further, choose cluster centers in an effective manner. Finally, Gower distance is used to cluster the data based on record similarity. Validation of the proposed clustering method was done using the two medical datasets, the Parkinson’s Disease (PD) and Wisconsin Diagnostic Breast Cancer (WDBC), and Bonn EEG datasets. Results obtained show drastic improvement in machine learning classifiers’ performance, achieving an accuracy of almost 100

This work introduces an efficient feature selection approach based on the Kulczynski Similarity (KS) measure to improve the efficiency and accuracy of machine learning models in Electroencephalogram (EEG)-based autism spectrum disorder (ASD) detection. EEG data are inherently high-dimensional, and using all features increases computational burden while incorporating redundant or irrelevant attributes that may hinder classification performance. The proposed KS method addresses these issues by ranking features according to their similarity, retaining those with high harmony that are most relevant for diagnosis, and discarding non-contributory dimensions. To evaluate the effectiveness of the KS framework, three configurations were compared: the complete EEG feature set, features selected with standard Particle Swarm Optimization (PSO), and features selected with the proposed KS method. While PSO retained 30 out of 47 features (63.83

This paper presents an efficient feature selection based on Ruzicka similarity to detect and diagnoses seizures caused by epilepsy. The proposed approach reduces the feature space while retaining the most relevant features for classification, enhancing the performance of standard Machine Learning (ML) classifiers. Technically, Bonn University EEG dataset is utilized to validate the model, and classifiers such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree (DT), Naive Bayes (NB), and Random Forest (RF) are applied. Several measures such as accuracy, recall, precision, and F1-score have been applied for the model evaluation. Results demonstrate that the proposed Ruzicka-based feature selection method achieves superior classification accuracy of 100