Abstract:
In recent times, an upsurge of highly sophisticated and intricate malware has emerged, becoming one of the most insidious
and perilous attack techniques targeting critical information technology infrastructures. Android, the widely anticipated and open-source
smartphone operating system has experienced exponential growth. However, this progress has been impeded by the escalating threat of
Android malware, which exploits smartphones to carry out malicious acts. Malware employs a plethora of techniques to circumvent
detection systems, presenting novel obstacles to reliable detection. Detecting Android malware efficiently and accurately is crucial in
ensuring the security of Android OS users. Machine learning techniques have been widely employed to address this problem, and
feature selection algorithms have been introduced to enhance the detection process. This paper investigates the impact of feature
selection algorithms specifically applied to permission and API method information in detecting Android malware using different
machine learning algorithms. Experiments were conducted to compare the performance of feature selection algorithms, focusing on
Principal Component Analysis (PCA) feature selection, F-Score, Recursive feature selection, and Stochastic Neighbor Embedding
(SNE). The results demonstrate the effectiveness of the PCA algorithm-based approach in selecting relevant features for malware
detection, showing advantages over all feature selection algorithms and reducing the model-building time significantly. The findings
highlight the importance of feature selection in optimizing the machine learning-based malware detection system. By selecting pertinent
features, the detection process becomes more efficient, improving both accuracy and speed. The PCA algorithm-based feature selection
approach outperformed the Feature selection method, showcasing its ability to effectively identify features relevant to Android malware
detection.