Empirical Analysis of Feature Selection Techniques for Optimal Feature Set Identification in Android Malware Detection

Android is facing an exponential surge in malware, and legacy malware detection systems are ineffective in detecting new malware. Machine learning techniques aid in developing more effective malware detection systems, but high-dimensional datasets can pose a challenge. Feature selection is the process of pinpointing and choosing the most relevant features from a dataset to enhance model performance and decrease computation time. Filter methods, regardless of machine learning models, are effective for feature selection. This article reviews different filter methods, compares their predictive accuracy, and provides practical recommendations for using feature selection in malware detection. An empirical evaluation is presented on the selected Drebin dataset, comparing common filter methods selected from a detailed literature review. Various machine learning algorithms receive feature subsets selected from diverse filter methods to assess their performance. ReliefF outperformed other methods, random forest achieving the highest accuracy.