Dynamic analysis based Android Malware Detection using ANN and RFE Feature Selection

In recent years, extensive usage of mobile phones has led to a drastic rise in malware attacks. The classification of these malware has become extremely challenging as they are hidden in normal applications. However, technological growth in machine learning models makes the task of detecting existing malware and predicting the unknown malware possible. This paper proposes an Artificial Neural Network based dynamic malware analysis technique with CICMalDroid2020 dataset. The performance of ANN is improved in predicting unknown malware by applying different optimizers of ANN which minimize the error to 0.01%. All the input neurons (features of the dataset) of ANN may not contain relevant information. Hence, Recursive Feature Elimination technique is applied to identify features which are capable of classifying and predicting different categories of malware in the dataset. Experiment results show that only 34% of the total features can achieve around 99% accuracy to detect benign and malicious malware.