Explaining Android Application Authorship Attribution Based on Source Code Analysis

摘要

Source code authorship attribution is a process of source code authorship identification based on set of known code samples belonging to the given author. One of practical applications of code attribution is a malware analysis and detection. In the paper we explore attribution of Android applications based on classification of source code data with particular focus on explanation of the role of selected features and their impact on the final classifier decision. The proposed solution uses Local Interpretable Model-Agnostic Explanations (LIME) technique to explain decisions produced by classifiers. We explored this approach on several types of classifiers such as SVM, Random Forrest and neural network and dataset containing applications belonging to more than 20 different authors.