Machine Learning‐Driven Security for Malware Detection in Wireless Android Devices

ABSTRACT The persistent growth of Android malware has necessitated the development of advanced detection techniques to protect users and devices. This work proposes a machine learning method based on API calls and permissions to discriminate between malicious and harmless applications. The primary intent is to advance Android security by offering a comprehensible approach for detecting potentially dangerous apps before end users download them. A dataset of 11 000 applications (benign and malicious) is collected from the Androzoo Android apps collection. The binary dataset constructed from the extracted API calls and permissions has high dimensionality and sparsity, so a hybrid feature selection pipeline has been followed to filter out redundant, irrelevant, and low‐impact features while preserving the differential signs. The proposed method provides a scalable solution for risk assessment by differentiating between safe and risky apps with adequate outcomes. Experimental results show that combining permissions and API calls achieved an accuracy of 98.36% for classification, which is significantly higher than using either permissions or API calls alone.