Xddroid: Exploring Explainable Ai for Dynamic Analysis of Android Malware

With the ever-growing use of smartphones, there exists a greater chance of malicious attacks on Android-based devices, leaving the user's private and personal information at constant risk. Attackers target naive users who install Android applications without paying attention to what permissions are requested to install the application or what these permissions signify. Malicious applications often solicit more permissions than required, giving attackers access to sensitive information. In this paper, we have proposed XDDroid, an Android malware detection system based on dynamic analysis that integrates eXplainable AI to enhance the interpretability of the results. Additionally, we used the mutual information feature selection method to reduce the training time without impacting the accuracy. Based on how frequently applications make system calls, machine learning classifiers categorize them as benign or malicious. However, the model predictions appear as black boxes to the end user. So, to provide users with some explainability in identifying malware, we have used SHAP in XDDroid to explain the features responsible for classifying the application as benign or malicious.