Hyperparameter Tuning Strategies for Robust Machine Learning in Android Malware Detection

ABSTRACT The exponential growth of Android applications has been accompanied by a sharp rise in mobile malware, calling for intelligent and adaptive detection mechanisms to ensure user security. This research presents a robust framework for Android malware detection, integrating both static and dynamic analysis techniques. The proposed approach leverages machine learning models and hyperparameter tuning trained on a diverse dataset to enhance detection accuracy and adaptability to evolving threats. This framework is well‐suited for real‐time deployment in mobile security apps and app store screening systems, where timely and accurate threat detection is critical. To solve class imbalance, the random oversampling method was implemented to balance class distribution while improving the generalizability of the model. The models have been trained on the Drebin dataset, which has 15 036 (5560 malware and 9476 benign samples) from 179 different malware families. These samples, collected from August 2010 to October 2012 and contributed by the MobileSandbox project, improve detection accuracy and responsiveness to changing threats. The framework demonstrates strong classification performance with 98.67% accuracy, 99% precision, 99% recall, 99% F1‐score, an AUC‐ROC of 1.0, and a low False Positive Rate (FPR). These results underline the model's robustness and effectiveness in accurately identifying both known and emerging threats, making it well‐suited for real‐world deployment.