Active Learning-Based Mobile Malware Detection Utilizing Auto-Labeling and Data Drift Detection

Machine learning-based detection methods have demonstrated high performance in mobile malware detection. However, changes in mobile malware over time induce a sig-nificant challenge for malware detection systems running in operational environments. The development of non-stationary models to address concept drift in the threat landscape has garnered significant research interest. Auto-labeling is using automated algorithms to assign labels to data without manual intervention. This study explores the method for introducing auto-labeling in active learning with data drift detection. It achieves satisfying results for the lowest possible cost while succeeding in adapting to changes in the data for a long period.