<![CDATA[Malware poses a serious threat to Android security systems. In recent years, Android malware has rapidly evolved, employing obfuscation techniques such as polymorphic and metamorphic. Unfortunately, signature-based malware detection cannot identify modern variants of Android malware. This study aims to compare various feature selection methods and machine learning algorithms to identify the most effective and efficient combination for classifying Android malware. The dataset used in this research is the Drebin dataset. Four classification algorithms are used in this comparison: Naive Bayes, Logistic Regression, Neural Network, and Random Forest. The best-performing algorithm is then implemented in three different scenarios: without feature selection, with Information Gain, and with Chi-Squared (X²). In the latter two scenarios, the appropriate number of features was selected using the backward elimination method. Both feature selections achieved the same performance, but Information Gain required fewer features. The evaluation metrics used in this study include AUC, accuracy, F1-score, training time, and testing time. Measuring training and testing time benefits the model by making it more efficient, thus allowing for faster detection in real-world applications. The results show that the combination of the Information Gain feature selection method and the Neural Network algorithm achieves the highest performance, with an accuracy and F1-Score of 98.6%. Additionally, this combination achieves a training time of 81.135 seconds and a testing time of 1.095 seconds. Compared to the Neural Network algorithm without feature selection, this combination results in a 17.7597 % reduction in training time and a 57.9977 % reduction in testing time while maintaining the same performance values. This research contributes to improving the speed and accuracy of malware detection systems, enhancing mobile security.]]>
Copyrights © 2024
