Synthetic Minority Oversampling Technique Research Articles

Impact of imbalanced datasets on ML algorithms for malware classification

ABSTRACT This paper discusses the impact of imbalanced datasets on ML models for malware classification and whether disproportionate distribution of various families affects the ability of an ML model to learn from minority classes. Classification results learnt from ML model trained on an imbalanced dataset were compared against those trained on a dataset balanced using Synthetic Minority Oversampling Technique (SMOTE) and Tomek Links. Four machine learning models used included RF, SVM, DT, and KNN. The comparative measures used to evaluate models were accuracy, precision, and recall. Using the balanced dataset improved precision and recall for minority malware families, with the implementation of SMOTE and Tomek Links leading to measurable performance improvements across most classifiers. For example, RF accuracy for detecting Trojans rose from 95.4% to 97.6%, demonstrating the benefits of removing noisy samples to refine decision boundaries. Although SVM accuracy declined from 93.55% to 84.63%, improved precision and recall across other classifiers highlight that balancing techniques enhanced models’ ability to classify minority samples, reducing misleading high accuracies caused by class imbalance.

Characterisation of cardiovascular disease (CVD) incidence and machine learning risk prediction in middle-aged and elderly populations: data from the China health and retirement longitudinal study (CHARLS)

BackgroundDue to the ageing population and evolving lifestyles occurring in China, middle-aged and elderly populations have become high-risk groups for cardiovascular disease (CVD). The aim of this study was to analyse the incidence characteristics of CVD in these populations and develop a prediction model by using data from the China Health and Retirement Longitudinal Study (CHARLS).MethodsWe used follow-up data from the CHARLS to analyse CVD incidence in the Chinese middle-aged and elderly population over a time span of 9 years. Five machine learning (ML) algorithms were employed for risk prediction. Data preprocessing included missing value imputation via random forest. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (Lasso CV) method with cross-validation prior to model training. The application of the synthetic minority over-sampling technique (SMOTE) to address class imbalance. Model performance was evaluated via analyses including the area under the ROC curve (AUC), precision, recall, F1 score, and SHAP plots for interpretability.ResultsIn accordance with the exclusion criteria, 12,580, 12,061, 11,545, and 11,619 participants were enrolled in four follow-up rounds. The cumulative incidence (CI) of CVD at 2, 4, 7, and 9 years was 2.846%, 8.971%, 17.869% and 20.518%,, respectively. Significant differences in CVD incidence were observed across gender, age, ethnicity, and region, with higher rates observed in females and in the northeast region. Ultimately, 8,080 participants and 24 features were analysed for CVD risk prediction. Five ML models were built based on these features. Although the LGB model achieves an AUC of 0.818, indicating strong overall performance, its F1 score and recall rate are relatively low, at 0.509 and 43.1%, respectively. Shapley additive explanations (SHAP) analyses revealed the importance of key features, such as night sleep duration, TG levels, and waist circumference, in predicting outcomes, and highlighted the nonlinear relationships between these features and CVD risk.ConclusionsGender, age, ethnicity, and region are significant factors influencing CVD incidence. Although the LGB model demonstrates good overall performance, its low F1 score and recall rate reveal limitations in identifying high-risk cardiovascular disease patients.

Development of a machine learning model related to explore the association between heavy metal exposure and alveolar bone loss among US adults utilizing SHAP: a study based on NHANES 2015–2018

BackgroundAlveolar bone loss (ABL) is common in modern society. Heavy metal exposure is usually considered to be a risk factor for ABL. Some studies revealed a positive trend found between urinary heavy metals and periodontitis using multiple logistic regression and Bayesian kernel machine regression. Overfitting using kernel function, long calculation period, the definition of prior distribution and lack of rank of heavy metal will affect the performance of the statistical model. Optimal model on this topic still remains controversy. This study aimed: (1) to develop an algorithm for exploring the association between heavy metal exposure and ABL; (2) filter the actual causal variables and investigate how heavy metals were associated with ABL; and (3) identify the potential risk factors for ABL.MethodsData were collected from National Health and Nutrition Examination Survey (NHANES) between 2015 and 2018 to develop a machine learning (ML) model. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO) regression with 10-fold cross-validation. The selected data were balanced using the Synthetic Minority Oversampling Technique (SMOTE) and divided into a training set and testing set at a 3:1 ratio. Logistic Regression (LR), Support Vector Machines (SVM), Random Forest (RF), K-Nearest Neighbor (KNN), Decision Tree (DT), and XGboost were used to construct the ML model. Accuracy, Area Under the Receiver Operating Characteristic Curve (AUC), Precision, Recall, and F1 score were used to select the optimal model for further analysis. The contribution of the variables to the ML model was explained using the Shapley Additive Explanations (SHAP) method.ResultsRF showed the best performance in exploring the association between heavy metal exposure and ABL, with an AUC (0.88), accuracy (0.78), precision (0.76), recall (0.83), and F1 score (0.79). Age was the most important factor in the ML model (mean| SHAP value| = 0.09), and Cd was the primary contributor. Sex had little effect on the ML model contribution.ConclusionIn this study, RF showed superior performance compared with the other five algorithms. Among the 12 heavy metals, Cd was the most important factor in the ML model. The relationship of Co & Pb and ABL are weaker than that of Cd. Among all the independent variables, age was considered the most important factor for this model. As for PIR, low-income participants present association with ABL. Mexican American and Non-Hispanic White show low association with ABL compared to Non-Hispanic Black and other races. Gender feature demonstrates a weak association with ABL. In the future, more advanced algorithms should be developed to validate these results and related parameters can be tuned to improve the accuracy of the model.Clinical trial numbernot applicable.

A novel aggregated coefficient ranking based feature selection strategy for enhancing the diagnosis of breast cancer classification using machine learning.

Effective Breast cancer (BC) analysis is crucial for early prognosis, controlling cancer recurrence, timely medical intervention, and determining appropriate treatment procedures. Additionally, it plays a significant role in optimizing mortality rates among women with breast cancer and increasing the average lifespan of patients. This can be achieved by performing effective critical feature analysis of the BC by picking superlative features through significant ranking-based Feature Selection (FS). Various authors have developed strategies relying on single FS, but this approach may not yield excellent results and could lead to various consequences, including time and storage complexity issues, inaccurate results, poor decision-making, and difficult interpretation of models. Therefore, critical data analysis can facilitate the development of a robust ranking methodology for effective feature selection. To solve these problems, this paper suggests a new method called Aggregated Coefficient Ranking-based Feature Selection (ACRFS), which is based on tri chracteristic behavioral criteria. This strategy aims to significantly improve the ranking for an effective Attribute Subset Selection (ASSS). The proposed method utilized computational problem solvers such as chi-square, mutual information, correlation, and rank-dense methods. The work implemented the introduced methodology using Wisconsin-based breast cancer data and applied the Synthetic Minority Oversampling Technique (SMOTE) to the obtained data subset. Later, we employed models such as decision trees, support vector machines, k-nearest neighbors, random forests, stochastic gradient descent, and Gaussian naive bayes to determine the type of cancer. The classification metrics such as accuracy, precision, recall, F1 score, kappa score, and Matthews coefficient were utilized to evaluate the effectiveness of the suggested ACRFS approach. The proposed method has demonstrated superior outcomes with fewer features and a minimal time complexity.

A Deep Learning CNN-GRU-RNN Model for Sustainable Development Prediction in Al-Kharj City

This study introduces an advanced Deep Learning (DL) framework, the Convolutional Neural Network-Gated Recurrent Unit-Recurrent Neural Network (CNN-GRU-RNN). This model is engineered to forecast climate dynamics extending to the year 2050, with a particular focus on four pivotal scenarios: temperature, air temperature dew point, visibility distance, and atmospheric sea level pressure, specifically in Al-Kharj City, Saudi Arabia. To address the data imbalance problem, the Synthetic Minority Over-Sampling Technique was employed for Regression along with the Gaussian Noise (SMOGN). The efficacy of the CNN-GRU-RNN model was benchmarked against five regression models: the Decision Tree Regressor (DTR), the Random Forest Regressor (RFR), the Extra Trees Regressor (ETR), the Bayesian Ridge Regressor (BRR), and the K-Nearest Neighbors Regressor (KNNR). The models were evaluated using five distinct metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error (MedAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2). The experimental outcomes demonstrated the superiority of the CNN-GRU-RNN model, which surpassed the traditional regression models across all four scenarios.

Optimized Random Forest Models for Rock Mass Classification in Tunnel Construction

The accurate prediction of rock mass quality ahead of the tunnel face is crucial for optimizing tunnel construction strategies, enhancing safety, and reducing geological risks. This study developed three hybrid models using random forest (RF) optimized by moth-flame optimization (MFO), gray wolf optimizer (GWO), and Bayesian optimization (BO) algorithms to classify the surrounding rock in real time during tunnel boring machine (TBM) operations. A dataset with 544 TBM tunneling samples included key parameters such as thrust force per cutter (TFC), revolutions per minute (RPM), penetration rate (PR), advance rate (AR), penetration per revolution (PRev), and field penetration index (FPI), with rock classification based on the Rock Mass Rating (RMR) method. To address the class imbalance, the Borderline Synthetic Minority Over-Sampling Technique was applied. Performance assessments revealed the MFO-RF model’s superior performance, with training and testing accuracies of 0.992 and 0.927, respectively, and key predictors identified as PR, AR, and RPM. Additional validation using 91 data sets confirmed the reliability of the MFO-RF model on unseen data, achieving an accuracy of 0.879. A graphical user interface was also developed, enabling field engineers and technicians to make instant and reliable rock classification predictions, greatly supporting safe tunnel construction and operational efficiency. These models contribute valuable tools for real-time, data-driven decision-making in tunneling projects.

Smartphone pupillometry with machine learning differentiates ischemic from hemorrhagic stroke: A pilot study.

Similarities between acute ischemic and hemorrhagic stroke make diagnosis and triage challenging. We studied a smartphone-based quantitative pupillometer for differentiation of acute ischemic and hemorrhagic stroke. Stroke patients were recruited prior to surgical or interventional treatment. Smartphone pupillometry was used to quantify components of the pupillary light reflex (PLR). A synthetic minority oversampling technique (SMOTE) was applied to correct sample size imbalance. Four binary classification model types were trained using all possible combinations of the PLR components with 10-fold cross validation stratified by cohort. Models were evaluated for accuracy, sensitivity, specificity, area under the curve (AUC), and F1 score. The three best-performing models were selected based on AUC. Shapley additive explanation plots were produced to explain PLR parameter impacts on model predictions. Eleven subjects with intraparenchymal hemorrhage and 22 subjects with acute ischemic stroke were enrolled. One way ANOVA demonstrated significant differences between healthy control data, AIS, and IPH in five out of seven PLR parameters. After SMOTE, each class had n=22 PLR recordings for model training. The best-performing model was random forest using a combination of latency, mean and maximum constriction velocity, and mean dilation velocity to discriminate between stroke types with 91.5% (95% confidence interval: 84.1-98.9) accuracy, 90% (82.9-97.1) sensitivity, 93.3% (83-100) specificity, 0.917 (0.847-0.987) AUC, and 90.7% (84.1-97.3) F1 score. Smartphone-based quantitative pupillometry could be useful in differentiating between acute ischemic and hemorrhagic stroke.

An intelligent computing methodology for two-phase flow performance assessment of electrical submersible pump using artificial neural network and synthetic minority over-sampling technique

An intelligent computing methodology for two-phase flow performance assessment of electrical submersible pump using artificial neural network and synthetic minority over-sampling technique

PTSP-BERT: Predict the thermal stability of proteins using sequence-based bidirectional representations from transformer-embedded features.

Thermophilic proteins, mesophiles proteins and psychrophilic proteins have wide industrial applications, as enzymes with different optimal temperatures are often needed for different purposes. Convenient methods are needed to determine the optimal temperatures for proteins; however, laboratory methods for this purpose are time-consuming and laborious, and existing machine learning methods can only perform binary classification of thermophilic and non-thermophilic proteins, or psychrophilic and non-psychrophilic proteins. Here, we developed a deep learning model, PSTP-BERT, based on protein sequences that can directly perform Three classes identification of thermophilic, mesophilic, and psychrophilic proteins. By comparing BERT-bfd with other deep learning models using five-fold cross-validation, we found that BERT-bfd-extracted features achieved the highest accuracy under six classifiers. Furthermore, to improve the model's accuracy, we used SMOTE (synthetic minority oversampling technique) to balance the dataset and light gradient-boosting machine to rank BERT-bfd-extracted features according to their weights. We obtained the best-performing model with five-fold cross-validation accuracy of 89.59% and independent test accuracy of 85.42%. The performance of the PSTP-BERT is significantly better than that of existing models in Three classes identification task. In order to compare with previous binary classification models, we used PSTP-BERT to perform binary classification tasks of thermophilic and non-thermophilic protein, and psychrophilic and non-psychrophilic protein on an independent test set. PSTP-BERT achieved the highest accuracy on both binary classification tasks, with an accuracy of 93.33% for thermophilic protein binary classification and 88.33% for psychrophilic protein binary classification. The accuracy of the independent test of the model can reach between 89.8% and 92.9% after training and optimization of the training set with different sequence similarities, and the prediction accuracy of the new data can exceed 97%. For the convenience of future researchers to use and reference, we have uploaded source code of PSTP-BERT to GitHub.

Counterfactual synthetic minority oversampling technique: solving healthcare’s imbalanced learning challenge

Counterfactual synthetic minority oversampling technique: solving healthcare’s imbalanced learning challenge

Assessing the environmental determinants of micropollutant contamination in streams using explainable machine learning and network analysis.

Even at trace concentrations, micropollutants, including pesticides and pharmaceuticals, pose considerable ecological risks, and the increasing presence of synthetic chemical substances in aquatic systems has emerged as a growing concern. Moreover, limited machine-learning (ML) approaches exist for analyzing environmental data, and the increasing complexity of ML models has made it challenging to understand predictor-outcome relationships. In particular, understanding complex interactions among multiple variables remains challenging. This study applies and integrates explainable ML techniques and network analysis to identify the sources of micropollutants in a large watershed and determine the factors affecting micropollutant levels. We assessed the performance of four ML algorithms-support vector machine, random forest, extreme gradient boosting (XGB), and autoencoder-XGB-in predicting micropollutant levels based on the spatial characteristics of the watershed. We applied the synthetic minority oversampling technique to address the data imbalance. The XGB model demonstrated superior predictive performance, particularly for high concentration levels, achieving an accuracy of 87%-99%. Shapley additive explanations (SHAP) analysis identified temperature and rainfall as significant factors. Moreover, agricultural activities contributed to pesticide pollution, whereas urban activities contributed to pharmaceutical contamination. The network analysis corroborated the SHAP findings and revealed event-specific contamination characteristics. This included distinct discharge pathways during a dry summer event and shared pathways during a wet winter event. This approach enhances an understanding of contamination sources and pathways and subsequently aids in developing control measures and making informed policy decisions to preserve water quality in mixed land-use areas.

Comparative study of imputation strategies to improve the sarcopenia prediction task.

Sarcopenia, a condition characterized by the progressive loss of skeletal muscle mass and strength, poses significant challenges in research due to missing data. Incomplete datasets undermine the accuracy and reliability of studies, necessitating effective imputation techniques. This study conducts a comparative analysis of three advanced methods-multiple imputation by chained equations (MICE), support vector regression, and K-nearest neighbors (KNN)-to address data completeness issues in sarcopenia research. Following imputation, we utilized machine learning models, including logistic regression, gradient boosting, support vector machine, and random forest, to classify sarcopenia. The methodology encompassed rigorous data preprocessing, normalization, and the synthetic minority oversampling technique to address class imbalance and ensure unbiased model performance. The results revealed substantial variations in model accuracy based on the imputation method employed. The gradient boosting model consistently exhibited superior performance across all imputation strategies, demonstrating its robustness with imputed datasets. Additionally, KNN and MICE emerged as effective imputation techniques, preserving the original data distribution and enabling more accurate classification outcomes. This study underscores the pivotal role of imputation methods in maintaining data integrity and enhancing predictive accuracy in sarcopenia research. The gradient boosting model's reliability across all strategies highlights its potential as a robust classifier, while the suitability of KNN and MICE for preserving data distribution supports their application in similar research contexts. These findings contribute to more reliable and valid insights in sarcopenia studies, ultimately supporting improved clinical outcomes.

Hyperparameter Tuning of Random Forest using Social Group Optimization Algorithm for Credit Card Fraud Detection in Banking Data

As the adoption of credit cards continues to expand alongside advancements in e-commerce, the frequency and complexity of fraudulent activities have also grown, posing significant challenges for the financial sector. Detecting fraudulent transactions within highly imbalanced datasets remains a critical issue in ensuring secure banking operations. This study explores a robust approach RF_SGO to credit card fraud detection by combining pre-processing techniques such as Synthetic Minority Oversampling Technique (SMOTE) and class weight adjustment with Random Forest (RF) models optimized using the Social Group Optimization (SGO) algorithm. Additionally, the study utilizes Random Forest's feature importance mechanism to identify the most influential features contributing to fraud detection, enhancing interpretability and decision-making. Our methodology evaluates RF_SGO across three datasets: the original European cardholders' imbalanced dataset, a class-weight-adjusted dataset, and a SMOTE-enhanced dataset. Model performance is measured using key metrics, including Accuracy, Precision, Recall, F1-Score, and ROC-AUC. The RF_SGO model demonstrated superior performance, with the SMOTE-enhanced variant achieving the highest ROC-AUC (0.98) and Recall (0.88), effectively balancing sensitivity and specificity. The class-weighted RF_SGO achieved the highest Precision (0.96), making it ideal for minimizing false positives. Furthermore, the feature importance analysis identified key predictors of fraudulent behavior, providing actionable insights for financial institutions. Comparisons with traditional machine learning algorithms (e.g., Logistic Regression, Decision Trees, and SVM) and advanced models (e.g., XGBoost, CatBoost, and deep learning) highlight RF_SGO's ability to outperform in precision-recall trade-offs and overall classification effectiveness. This study underscores the significance of incorporating hyperparameter tuning, feature importance analysis, and data balancing strategies to improve fraud detection. The proposed RF_SGO framework offers a scalable and efficient solution for financial institutions to mitigate fraud, ensuring more reliable and secure transaction systems.

Tunnel squeezing prediction based on partially missing dataset and optimized machine learning models

Accurate prediction of tunnel squeezing, one of the common geological hazards during tunnel construction, is of great significance for ensuring construction safety and reducing economic losses. To achieve precise prediction of tunnel squeezing, this study constructed six reliable machine learning (ML) classification models for this purpose, including Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), and K-Nearest Neighbors (KNN). The parameters of these 6 ML models were optimized using the Whale Optimization Algorithm (WOA) in conjunction with five-fold cross-validation. A total of 305 tunnel squeezing sample data were collected to train and test the models. KNN and Synthetic Minority Over-sampling Technique (SMOTE) methods were employed to handle the missing and imbalanced data sets. An input feature system for tunnel squeezing prediction was established, comprising tunnel burial depth (H), tunnel diameter (D), strength-to-stress ratio (SSR), and support stiffness (K). The XGBoost model optimized with WOA demonstrated the highest prediction accuracy of 0.9681. The SHAP method was utilized to interpret the XGBoost model, indicating that the contribution rank of the input features to tunnel squeezing prediction was SSR > K > D > H, with average SHAP values of 2.93, 1.49, 0.82, and 0.69, respectively. The XGBoost model was applied to predict tunnel squeezing in 10 sections of the Qinghai Huzhu Beishan Tunnel. The prediction results were highly consistent with the actual outcomes.

Enhancing Mental Health Disorder Detection: A Hybrid Classifier System Using Soft Voting and Ensemble Methods

Early detection and diagnosis of mental health disorders based on patient-reported symptoms is vital in mental health care. This research proposes an ensemble learning pipeline for predicting mental health disorders using symptoms and demographic data. The pipeline incorporates preprocessing techniques such as standardization, one-hot encoding, and Term Frequency - Inverse Document Frequency transformation, and addresses class imbalance with the Synthetic Minority Over-sampling Technique to enhance predictions of rare disorders. Multiple classifiers, including Random Forest, Gradient Boosting, XGBoost, CatBoost, and LightGBM, are combined in a voting classifier, with each model optimized using Grid Search. The pipeline's performance is evaluated on its ability to predict mental health disorders, showing potential for supporting clinical diagnosis. CatBoost proves effective in certain models, while LightGBM delivers strong performance across most. The models achieve around 95% accuracy, demonstrating strong discrimination between disorders and other conditions.

IMPLEMENTATION OF THE PSO-SMOTE METHOD ON THE NAIVE BAYES ALGORITHM TO ADDRESS CLASS IMBALANCE IN LANDSLIDE DISASTER DATA

Landslides are a significant natural disaster threat in Indonesia, including Samarinda City. This study aims to evaluate the classification accuracy of landslides using the Naïve Bayes algorithm, combined with the Synthetic Minority Oversampling Technique (SMOTE) for class imbalance and Particle Swarm Optimization (PSO) for feature Selection and parameter optimization. Data were obtained from BPBD and BMKG Samarinda for 2022-2024, evaluated using 10-Fold Cross Validation and Confusion Matrix. The results show that SMOTE did not always improve model accuracy. Naïve Bayes without SMOTE achieved an average accuracy of 87.18%, while with SMOTE, it decreased to 70.55%. This indicates that SMOTE’s impact depends on the dataset. PSO, used for feature Selection and parameter optimization, gave mixed results. Naïve Bayes + PSO Feature Selection achieved an average accuracy of 68.25%, lower than the base model with SMOTE. Naïve Bayes + PSO Optimization achieved the best accuracy at 71.03%, though the improvement was minimal. The combination of Naïve Bayes + PSO Feature Selection + PSO Optimization had the lowest accuracy at 68.08%. This study emphasizes that the use of PSO for feature Selection and optimization needs careful consideration, especially when combined with SMOTE, as its effect on model performance is not always optimal.

Advanced sleep disorder detection using multi-layered ensemble learning and advanced data balancing techniques

Sleep disorder detection has greatly improved with the integration of machine learning, offering enhanced accuracy and effectiveness. However, the labor-intensive nature of diagnosis still presents challenges. To address these, we propose a novel coordination model aimed at improving detection accuracy and reliability through a multi-model ensemble approach. The proposed method employs a multi-layered ensemble model, starting with the careful selection of N models to capture essential features. Techniques such as thresholding, predictive scoring, and the conversion of Softmax labels into multidimensional feature vectors improve interpretability. Ensemble methods like voting and stacking are used to ensure collaborative decision-making across models. Both the original dataset and one modified using the Synthetic Minority Oversampling Technique (SMOTE) were evaluated to address data imbalance issues. The ensemble model demonstrated superior performance, achieving 96.88% accuracy on the SMOTE-implemented dataset and 95.75% accuracy on the original dataset. Moreover, an eight-fold cross-validation yielded an impressive 99.5% accuracy, indicating the reliability of the model in handling unbalanced data and ensuring precise detection of sleep disorders. Compared to individual models, the proposed ensemble method significantly outperformed traditional models. The combination of models not only enhanced accuracy but also improved the system's ability to handle unbalanced data, a common limitation in traditional methods. This study marks a significant advancement in sleep disorder detection through the integration of innovative ensemble techniques. The proposed approach, combining multiple models and advanced interpretability methods, promises improved patient outcomes and greater diagnostic accuracy, paving the way for future applications in medical diagnostics.

Machine Learning Approaches for Effective Credit Card Fraud Detection: Addressing Imbalance and Enhancing Accuracy

Credit card fraud has emerged as a significant threat to the financial sector, driven by the rapid growth in online transactions and the evolving sophistication of fraudulent activities. This research aims to design and implement a machine learning-based solution capable of detecting fraudulent credit card transactions effectively. By addressing challenges such as dataset imbalance and false positives, the research employs preprocessing techniques including Synthetic Minority Oversampling Technique (SMOTE), along with advanced machine learning algorithms like Logistic Regression, XGBoost, and Isolation Forest. It highlights the potential of these models to enhance fraud detection accuracy and scalability, providing a practical and deployable tool for real-world applications.

Unlocking the power of optimized data balancing ratios: a new frontier in tackling imbalanced datasets

Data balancing methods eliminate the problem of imbalanced class distributions, which often lead to the majority class being well-learned while the minority class remains underrepresented, negatively affecting classification performance. This study applies data balancing to the healthcare domain, a critical field where classification success directly impacts human life. The primary aim is to introduce novel balancing methods while addressing the previously overlooked problem of optimizing data balancing ratios. Six healthcare datasets were used: Wisconsin Diagnostic Breast Cancer (WDBC), Wisconsin Prognostic Breast Cancer (WPBC), Z-Alizadeh Sani, Kidney, Diabetes, and Stroke, all characterized by significant diseases and imbalanced class distributions. Six balancing methods were tested, including synthetic minority oversampling technique (SMOTE), adaptive synthetic sampling (ADASYN), support vector machine-SMOTE (SVM-SMOTE), Borderline-SMOTE, cubic interpolation, and quadratic interpolation, with interpolation-based methods being adapted to this domain for the first time. The critical factor in data balancing is identifying the optimal ratio that maximizes classification performance. In this study, particle swarm optimization (PSO), whale optimization algorithm (WOA), and Optuna optimization methods were used to optimize balancing ratios via a custom-designed fitness function that simultaneously optimizes classification accuracy and resource consumption. Classification was conducted for three scenarios: full balance, optimized balance, and imbalance, using support vector machine (SVM), random forest (RF), and ensemble learning (EL) classifiers, allowing for extensive analysis. Each combination of balancing methods, classifiers, and optimization techniques was separately analyzed using metrics such as accuracy, precision, recall, F1-score, time, central processing unit (CPU) usage, and memory usage. As a result, the combination that optimally balances classification accuracy and resource consumption was determined for each dataset, providing both comprehensive analysis and insights into the impact of balancing ratio optimization on diagnostic success in health care.

Enhanced Hybrid Deep Learning Models-Based Anomaly Detection Method for Two-Stage Binary and Multi-Class Classification of Attacks in Intrusion Detection Systems

As security threats become more complex, the need for effective intrusion detection systems (IDSs) has grown. Traditional machine learning methods are limited by the need for extensive feature engineering and data preprocessing. To overcome this, we propose two enhanced hybrid deep learning models, an autoencoder–convolutional neural network (Autoencoder–CNN) and a transformer–deep neural network (Transformer–DNN). The Autoencoder reshapes network traffic data, addressing class imbalance, and the CNN performs precise classification. The transformer component extracts contextual features, which the DNN uses for accurate classification. Our approach utilizes an enhanced hybrid adaptive synthetic sampling–synthetic minority oversampling technique (ADASYN-SMOTE) for binary classification and enhanced SMOTE for multi-class classification, along with edited nearest neighbors (ENN) for further class imbalance handling. The models were designed to minimize false positives and negatives, improve real-time detection, and identify zero-day attacks. Evaluations based on the CICIDS2017 dataset showed 99.90% accuracy for Autoencoder–CNN and 99.92% for Transformer–DNN in binary classification, and 99.95% and 99.96% in multi-class classification, respectively. On the NF-BoT-IoT-v2 dataset, the Autoencoder–CNN achieved 99.98% in binary classification and 97.95% in multi-class classification, while the Transformer–DNN reached 99.98% and 97.90%, respectively. These results demonstrate the superior performance of the proposed models compared with traditional methods for handling diverse network attacks.