ENSIC: Feature Selection on Android Malware Detection Attributes Using an Enhanced Non-Linear SVM Integrated with Cross Validator

Abstract

The digital telecom world made the humans to have android based mobiles. Hackers are realising more number of threatening applications daily to attack the android devices either to steal the personal information or to commit fraud money transaction. These cyber crimes, which attacks the devices functionality or security is known as “Malware Attacks”. The proposed research aims to identify the impact of the android device factors like call signals, intents, and commands to access an application in classifying the malware attacks. Existing dataset contains 215 sub attributes for the given factors, which is treated as a high dimensionality according the principles of the machine learning. So, the previous researchers implemented traditional feature selection techniques like correlation, which consumes high computational time because bi-variate or multi variate with chi-square analysis needs to do lot of computations in between the different pairs of attributes. Using the recursive feature elimination has produced more number of attributes, which is more than half of the attributes and has got average accuracy with high mis-classification rate. The proposed research enhances the embedded approach by integrating the non-linear customized SVM (Support Vector Machine) with cross validated pipelining feature. This increases the accuracy of the model and reduces the number of essential attributes to 74 with 95.58% accuracy.