COMBATTING BANKING MALWARE THREATS: EVALUATING THE EFFICACY OF HYBRID AND SINGLE-CLASSIFICATION ALGORITHMS

Abstract

The increasing sophistication and prevalence of banking malware pose significant challenges to cybersecurity, threatening the confidentiality, integrity, and availability of financial systems and user data. This study evaluates the efficacy of hybrid and single-classification algorithms in detecting banking malware, addressing a critical gap in existing research. A total of eight classification algorithms were analyzed, including three hybrid models—Stacked Ensemble with Gradient Boosting, AdaBoost, and Stacking with Decision Trees and Random Forest. Additionally, five single classifiers—Support Vector Machine (SVM), Decision Tree, k-NN, Random Forest and Logistic Regression were assessed. The research methodology incorporated principal component analysis (PCA) for feature selection and techniques like Adasyn and Tomek Link to address data imbalance. Classification performance was evaluated using key metrics: accuracy, precision, recall, and F1-score. Results demonstrated that hybrid models, particularly an ensemble combining Random Forest and Decision Tree, outperformed other classifiers, achieving superior accuracy (0.98), precision, and recall. While Gradient Boosting and AdaBoost also exhibited robust performance, Logistic Regression showed room for improvement in precision and recall metrics. This research highlights the effectiveness of hybrid classification models in enhancing the detection of banking malware and underscores their potential for strengthening cybersecurity defenses in financial systems. The study contributes to the growing literature on machine learning applications in malware detection and provides insights into the strengths and limitations of diverse classification algorithms.