Authors: Nikhil Bhamare, Piyush Takalkar, Sujit Sherkar, Ms. Rajashri Malage
Abstract: The exponential growth of the Android ecosystem has been accompanied by a surge in sophisticated mobile mal-ware. Traditional signature-based detection mechanisms struggle to keep pace with these evasive threats, necessitating more adaptive and intelligent defense strategies. In this paper, we present a novel hierarchical ensemble Convolutional Neural Network (CNN) framework designed for robust Android malware detection. By transforming APK bytecode into grayscale images, our approach bypasses conventional manual feature engineering and leverages spatial pattern recognition. The proposed archi-tecture integrates three distinct deep learning models—ResNet50, DenseNet121, and VGG16—to extract diverse and comprehensive feature representations. The framework operates in two stages: initially classifying applications as benign or malicious, and subsequently categorizing the malicious samples into 25 distinct malware families. Experimental evaluations demonstrate that our ensemble approach achieves a high accuracy of 89.15%, out-performing individual CNN baselines. Furthermore, this image-based learning paradigm proves highly resilient to common structural obfuscation techniques utilized by modern Android malware.
DOI: http://doi.org/
