An Efficient Security Testing for Android Application Based on Behavior and Activities Using Improved PCA and DNN-KNN Classifier

Privacy and the safety of users are seriously threatened by Android, a mobile operating system that is becoming more and more popular. Due to the single feature selection technique's poor detection effects, the typical machine learning algorithm's ineffective detection efficiency, and the low accuracy caused by inappropriate feature extraction from log files. To overcome these issues, an efficient security testing method is designed based on the behaviour and activities of Android applications. In this designed model, efficient security testing is conducted by collecting the Android log file from the Android database. And the requested permissions are evaluated with the prioritized and categorized permissions for accessing the components of the Android device. If the permission threat level is low, the component permission will be allowed or else the permissions will be denied. The denied permissions are processed for security testing to predict the app status whether malware or benign. For predicting the state of the Android application, the log files are collected for extracting the features related to the behavior and activities using the improved PCA technique. For determining the number of components in the PCA based on the K-fold cross validation technique, the enhanced PCA technique combines PCA with the BES optimization technique. The hybrid DNN-KNN classifier is trained and evaluated with the best set of features from the enhanced PCA for identifying malicious and benign applications. Performance criteria including Accuracy, Recall, Specificity, and Precision are assessed for this created model. The attained performance metrics values for the proposed model are 98%, 97.6%, 99.2% and 97.7%. Using this proposed hybrid model malicious activity in android apps can be detected effectively and the public can be safeguared from serious threats.