ENHANCED ANDROID MALWARE DETECTION USING A DEEP LEARNING– DRIVEN ENSEMBLE FRAMEWORK

Abstract

With the rapid growth of smartphone usage in India, Android has become the dominant mobile operating system, accounting for over 95% of mobile devices. This popularity has also made Android a prime target for malware attacks. Reports indicate that India consistently ranks among the top countries affected by mobile malware, including trojans, spyware, ransomware, and adware. The objective of this work is to design a deep learning– based ensemble framework that accurately detects Android malware, classifies threat severity, and identifies attack families for enhanced mobile security. In manual or conventional systems, Android malware detection relies primarily on signature-based antivirus tools and rule-based analysis. Security experts manually analyze application code, permissions, and behaviors, then compare them against known malware signatures or predefined rules to determine whether an application is malicious or benign. Manual systems are ineffective against new and zero-day malware, as they depend heavily on predefined signatures and expert intervention. They are time-consuming, error-prone, unable to scale with large app volumes, and struggle to detect obfuscated or evolving malware variants, resulting in low detection accuracy. The increasing complexity, volume, and evolution of Android malware expose the limitations of manual and signature-based systems. This research is motivated by the need to overcome poor adaptability, delayed detection, and low accuracy by introducing intelligent models capable of learning complex malware patterns automatically and efficiently. The proposed system introduces a deep learning–based ensemble framework for robust Android malware detection. It integrates CNNs to capture spatial patterns from application features, LSTMs to model sequential and behavioral characteristics, and Random Forest classifiers to enhance decision stability. These models are combined using a Deep Ensemble Fusion Tao Tree mechanism to improve generalization and reliability. The framework detects malware type (benign or malicious), classifies threat severity, and identifies attack families such as Trojan, Spyware, Ransomware, and Adware. By leveraging ensemble learning and deep feature extraction, the system achieves higher accuracy, resilience to obfuscation, and effec