Machine Learning and Deep Learning Approaches for Android Malware Detection: An Empirical Evaluation

The proliferation of Android devices has significantly increased malware threats, compromising user privacy and system security. To address this, the study presents an empirical evaluation of machine learning (ML) and Deep Learning (DL) techniques for Android malware detection. The following models were evaluated: Least Squares Support Vector Machine (LS-SVM), Kernel Extreme Learning Machine (KELM), Regularised Random Vector Functional Link Neural Network (RRVFLN), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Random Forest, using the CICMalDroid-2020, CICAndMalDroid-2020, and Drebin datasets. LS-SVM achieved the highest accuracy of 98.9%, with strong precision and recall, while DL models demonstrated higher recall but lower generalisation. The proposed framework reduces false positives and offers a scalable solution for Android cybersecurity. These findings highlight the strengths and trade-offs between ML and DL methods. Future work includes enhancing adversarial robustness and enabling real-time detection, contributing to the development of more resilient malware detection systems for Android environments.