Revisiting Behavioral Android Malware Classification: A Leakage-Controlled Benchmark of Tree Ensemble Methods and Hybrid Attention Networks

Presents a leakage-controlled benchmark comparing strong tree ensembles with a hybrid attention network on behavioral Android malware data. The study shows that well-tuned tree ensembles outperform the more complex HAT-Net while maintaining strong accuracy and efficiency, providing a rigorous and reusable evaluation methodology for tabular malware detection