BEHAVIOUR ANALYSIS OF MOBILE SENSORS AND DETECTING UNUSUAL BEHAVIOUR

Abstract

Today smartphones become a vital part of every individual in the world and smartphones have tied the users with a strong bond for every day to day task like setting alarm, order food, making payments and many more. A user never knows what going on inside their phones. A user cannot detect a mobile application whether it is having any malicious behaviour by its appearance, say you have downloaded an application from play store or any third-party store and that app is transmitting your personal data to a remote server without your knowledge. Even google play store sometimes cannot detect these applications due to code obfuscation techniques. In this research, we are analysing the behaviour of mobile sensors in malicious and benign mode and we are trying to detect if any application performs any malicious activity based on our analysis. We are using sherlock dataset for the behavioural analysis and applied four supervised machine learning techniques to detect unusual behaviour and comparing the results to find which technique is most accurate. We have taken 2 feature sets first contains only application features and other contain global features with application features. We have used F1 score as a deciding parameter for best performance. In results, we have found that XGBoost performs best with F1 score of 98.82 and 98.86 on applications global dataset respectively.