A Novel Wavelet-Based Model For Android Malware Detection Utilizing System Calls Features

Abstract Security is a significant concern in the realm of Android platforms, given the ability to allow installations from unverified sources exposing devices to possible malware infiltration and malicious attacks. Malware classification remains a challenging task in the field of mobile malware detection, where dynamic analysis plays a central role in intrusion detection systems (IDS). This process aims real-time examination of application behavior enabling extraction of dynamic patterns exhibited by malware. System calls, provided by the device’s operating system and accessible from user applications, are deployed in machine learning methods for malware detection. The performance of machine learning-based detection algorithms requires improving the feature selection within system calls. This work suggests an original approach based on wavelets to enhance detection of a wide spectrum of Android malwares, such as adware, riskware, banking and SMS. The present methodology uses the Chi-Square test for feature selection, and Haar wavelet for converting selected attributes into wavelet coefficients. In our experiments, we use several machine-learning classifiers such as Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), and Neural Network (NN). To assess the performance of these classifiers, we employ evaluation metrics in terms of Accuracy, Recall, F-Score, and Precision. Among the selected classifiers, the Random Forest model coupled with wavelet feature selection provided the highest performance, achieving an accuracy rate of 99.99%. The results clearly demonstrate the effectiveness of our proposed model.