Abstract
Malicious software has increased massively on smartphones, tablets, etc. and has become popular all over the world in recent times. Multiple static and dynamic malware detection methods were proposed efficiently to detect Android malware. Permission-based/feature-based malware detection through API feature selection is one of the most popular techniques for detection. However, ignoring specific features or securing unrelated features may cause incredulity for classification algorithms with respect to classification time and improvement of accuracy. In order to face this challenge, different feature reduction tools were suggested by researchers to improve the accuracy of detection without considering the time. In this work, feature extraction, and feature ranking, are performed through PCA analysis and classification of apk’s attained better accuracy with less prediction time. This method involves the combination of malware and benign applications (19611 entries) and the list of ranked features. Among the 79 features in the dataset, the feature visualization of 11 features is shown for benign and malicious applications. Here, the random forest algorithm performs the classification with parameters like precision, f1-score, and recall. This model observed a 0.90 f1-score while accuracy is 96% and the weighted accuracy is 94%.
Keyword
Malicious software, android malware, API features, classification, Feature Ranking, Feature Extraction, PCA.
PDF Download (click here)
|