Classification of Malicious Android Applications Using Naive Bayes and Support Vector Machine Algorithms

As the use of smart devices increases, the number of malicious software is also increasing day by day. Android is the most used operating system in smart devices. That's why there is a lot of malware targeting this platform. By examining the permission properties of malicious software, it can be determined whether it is malicious or not. However, this is a complex problem. In order to solve this problem, in this study, classification processes have been carried out to determine whether the software is harmful with machine learning methods. For this purpose, a dataset containing 2854 malicious software and 2870 harmless software was created. In the dataset, there are 116 permission features for each software and a class feature that indicates whether it is malicious or not. Using these features, Support Vector Machine (SVM) and Naïve Bayes (NB) models were trained. The 10-fold cross validation method was used in training and testing processes. Accuracy, F-1 Score, precision, recall and specificity metrics were used to analyze the performances of the models. ROC curve and AUC values were used to analyze the learning and prediction levels of the models. As a result of the tests of the models, 90.9% classification success was obtained from the SVM model and 92.4% from the NB model.