Assessing the efficacy of Machine learning classifier for Android malware detection

Abstract

The primary challenges faced by software security experts is the identification and detection of malware within Android applications, as dangerous software is increasingly being embedded in sophisticated manners in application software. The existing applications, as well, are expanding in size and becoming increasingly intricate in terms of their functionalities. The ongoing endeavor of extracting valuable and indicative functionality from applications is a perpetual undertaking. There has been a lack of comprehensive studies that examine the specific attributes designed for identifying malicious applications on the Android platform. This is despite the existence of several feature extraction methods employed in prior research endeavors. Here, a comprehensive and concise analysis is presented to comprehend the behavior of applications using various criteria to identify harmful applications. This study evaluates the efficacy of ten different machine learning classifiers by analyzing a dataset including 15,036 applications categorized as either harmful or benign. The evaluation of classifiers involved the utilization of many metrics like Accuracy, Area Under the Curve (AUC), False Positive Rate (FPR), and False Negative Rate (FNR) towards development of illustrative framework for the detection of Android malware applications.

URN:NBN:sciencein.jist.2024.v12.788