A Hybrid Feature Selection Approach-Based Android Malware Detection Framework Using Machine Learning Techniques

With more  popularity and advancement in Internet-based services, the use of the Android smartphone has been increasing very rapidly. The tremendous popularity of using the Android operating system has attracted malware attacks on these devices. Detecting variants of malware features that change their behavior to hide from being detected by the traditional method of machine learning is being an incapable and challenging task. To overcome these issues of malware feature detection, an efficient feature selection plays a crucial role in detecting malware features and reduces the dimensionality of a huge dataset and removes the unnecessary features that are not useful and keeps those relevant features that improve the classification accuracy and detection rate. To address the above issues, this paper proposed a novel framework in which a hybrid feature selection using wrapping feature selection (WFS) with the combination of random forest and greedy stepwise (RF-GreedySW) framework is devised to optimize the malware features. The proposed framework is capable of reducing a large number of attributes into an optimal feature to enhance the performance of the machine learning model. The framework used the three most popular ML classifiers such as random forest (RF), decision tree (C5.0), and support vector machine radial basis function (SVM RBF). The performance of the proposed framework is evaluated using the CIC-InvesAndMal2019 dataset. The DT (C5.0), RF, and SVM RBF model achieves better accuracy of 91.80%, 91.32%, and 82.33% on static layer, respectively. Similarly, the accuracy is 72.41%, 75.10%, and 62.07% on the dynamic layer by DT (C5.0), RF, and SVM RBF, respectively. Our model highlights good results on the CIC-InvesAndMal2019 dataset in terms of classification accuracy and increases the robustness of the model.