Multi-feature image analysis for Android malware classification using convolutional neural networks
Pham Nhat Duy, Nguyen Tan Cam · 2024
Mobile security threats have increased due to the growing popularity of Android apps, especially in malware detection. Current methods give good performance in binary classification but most have difficulty in multi-class classification. This study introduces a novel approach to enhance multi-class classification of Android malware through image-based classification. The proposed system performs a conversion of Android application files into specialized images, including Markov, Entropy graph, and Gray-level matrix images. This set of images is used to train fine-tuned CNN (Convolutional Neural Network) models and it shows better performance than existing techniques. Experiments were conducted on the latest Android benchmark dataset CICMalDroid 2020. The method achieves an accuracy of 94.79% and 1.3% false positive rate.