Published Paper Details:

Adware, often dismissed as a low-priority threat in mobile security, has evolved into a sophisticated attack vector capable of compromising user privacy, draining resources, and facilitating data exfiltration. This paper presents a novel, behavior-based approach to detecting adware attacks on mobile phones, addressing the growing challenge of aggressive ad libraries, click fraud, and covert data transmission. Our methodology integrates both static and dynamic feature extraction, focusing on permission overuse, sensitive API call patterns, and behavioral anomalies such as excessive CPU usage, battery drain, and ad network interactions. We employ a Random Forest classifier optimized for mobile deployment, achieving a detection accuracy of 96.12% and an AUC-ROC score of 0.9914. Explainability techniques, including SHAP analysis, provide transparent insights into model decisions, highlighting key features like READ PHONE STATE and RECEIVE BOOT COMPLETED as primary indicators. The proposed system, with its lightweight architecture (1.7 MB) and low false positive rate (1.42%), offers practical applications for enterprise mobile device management, app store review automation, and end-user security tools. Future directions include incorporating network traffic analysis and adaptive learning to counter advanced adware evasion techniques.