A Model for Android Platform Malware Detection Utilizing Multiple Machine Learning Algorithms

In today's technological landscape, the ubiquitous use of mobile devices underscores their critical importance in facilitating daily tasks and enabling a wide array of functionalities, from communication to commerce and entertainment. However, this widespread adoption also brings significant concerns regarding security and privacy, especially with the proliferation of mobile applications capable of accessing sensitive data without explicit user consent. The Android operating system, renowned for its openness and extensive app ecosystem, faces substantial security challenges due to its susceptibility to malware attacks. Malicious software, covertly embedded within seemingly legitimate apps, poses serious threats such as data theft, unauthorized access, and device compromise. This study presents a comprehensive approach to malware detection on the Android platform, utilizing a dataset comprising 4,464 instances, evenly divided between 2,533 labeled as Malware and 1,931 labeled as Benign. The dataset, sourced from real-world Android applications, includes 328 extracted features to enhance detection accuracy. Five machine learning algorithms were evaluated to develop a robust malware detection system: Random Forest, Extra Trees, Logistic Regression, Gradient Boosting, and Support Vector Machine. The performance of these algorithms was rigorously assessed based on accuracy, precision, recall, F1-score, and ROC-AUC. The performance of these algorithms is rigorously evaluated and compared based on accuracy, precision, recall, and F1-score. The results reveal that the Logistic Regression algorithm achieved the highest accuracy at 97.31%, outperforming the other models. Specifically, Random Forest achieved 96.64%, Extra Trees 96.08%, Gradient Boosting 96.19%, and Support Vector Machine 96.75%. These findings suggest that Logistic Regression is particularly effective in identifying Android malware within this dataset, offering a reliable solution for enhancing mobile security. This research benchmarks these results against prior models utilizing different machine learning approaches and provides concrete insights into the most effective methodologies for mitigating Android malware threats. By advancing detection capabilities through sophisticated machine learning techniques, this study contributes to ongoing efforts to safeguard mobile device users from evolving cybersecurity threats, underscoring the critical role of data-driven models in enhancing the security and privacy of Android platforms. © 2024 Slovene Society Informatika. All rights reserved.