This paper presents a predictive analysis for detecting malware on Android devices using Machine Learning and explainability methods to interpret the results. After preprocessing, the dataset was reduced to 34,076 samples and 179 features of system calls and permissions. Among the 13 classifiers evaluated, eXtreme Gradient Boosting (XGBoost) proved to be the most efficient, with accuracy, precision, recall, and F1-Score metrics of approximately 94% and a training time of 1.48s. The SHapley Additive exPlanations (SHAP) method was used to explain the model’s decisions, which revealed system calls and sensitive permissions, such as READ PHONE STATE, SYSTEM ALERT WINDOW, SEND SMS, ACCESS WIFI STATE, getpriority, and getrlimit strongly associated with malwares.
|
