CURRENT ISSUES

ABSTRACT

Android malware continues to present a serious security challenge because harmful applications can alter permissions, program structure, API use, and behavior in ways that make conventional detection more difficult. This study presents an attention-enhanced deep-learning framework for Android malware detection and develops an existing undergraduate project into a research-oriented investigation. A labeled dataset containing 1,025 records and 14 features is used to compare a neural-network classifier with a Random Forest baseline. The original setup employed an 80:20 stratified division, giving 820 training records and 205 testing records. Both models reported 100% accuracy on the test set, whereas five-fold cross-validation of the Random Forest gave a mean accuracy of 99.41%. Since such high performance may also be affected by dataset characteristics or information leakage, the study stresses reproducibility, duplicate and leakage checks, family-aware evaluation where labels are available, and the use of precision, recall, F1-score, ROC-AUC, and false-positive rate. The proposed research direction combines attention, robustness, and explainability to support further investigation of changing and previously unseen Android malware. Keywords— Android malware detection; deep learning; attention mechanism; Random Forest; machine learning; mobile security; malware classification; explainability

Keywords: Android malware detection; deep learning; attention mechanism; Random Forest; machine learning; mobile security; malware classification; explainability