MH-AutoML: solução AutoML para o domínio de Malwares Android

Abstract

This study investigates the use of Automated Machine Learning frameworks (AutoML) and proposes the development of MH-AutoML (Malware Hunter Automated Machine Learning), a specialized framework designed for Android malware detection. This solution represents an evolution of the award-winning DroidAutoML, which was recognized as the second-best tool at the Brazilian Symposium on Information and Computational Systems Security (SBSeg 2022). The core of this study lies in simplifying the crucial processes inherent to the machine learning pipeline, including data cleaning, feature engineering, algorithm selection, hyperparameter tuning, and model evaluation. MH-AutoML is being meticulously developed with a special emphasis on customization, transparency, interpretability, and debugging. Additionally, principles of low coupling have been applied, substantially improving the tool's organization as well as its capacity for evolution and maintenance. The work concludes with discussions covering potential improvements and future directions for MH-AutoML. These considerations acknowledge the ongoing importance of refinement and innovation in this fundamental research domain.