Examining Malware Patterns in Android Platform using Sufficient Input Subset (SIS)

Smartphones are now inseparable part of our reality. Several machine learning algorithms exist for detection of malwares in android applications; however, these techniques fail to rationalize specific decisions made by a “Black Box” therefore lacking explain-ability. To overcome this limitation, Sufficient Input Subset (SIS) technique is used along with convolutional neural network (CNN). SIS categorizes minimal subsets of features who's observed values alone be sufficient for the same verdict to be reached. The results of the proposed technique are very promising., where its detection accuracy reached more than 90% and we are able to rationalize why the Black box classified a file as malware.