Addressing Class Imbalance in Malware Detection with Cost-Sensitive Learning: A Framework for Enhanced Performance

Abstract Malware detection remains a key yet challenging aspect of computer security. Mainly due to the constantly evolving nature of malware and the growing complexity of cyber-attacks. Conventional malware detection methods are known to perform poorly when confronted with imbalanced datasets. In this study, we introduce a cost-sensitive ensemble learning methodology combined with feature importance analysis to effectively address the class imbalance problem in malware detection tasks. Specifically, we applied different cost learning matrices on the Light Gradient Boosting Machine (LightGBM) and Random Forest (RF) as feature selection techniques to measure their impact on detection performance. Our results showed that cost-sensitive learning can improve malware detection accuracy on imbalanced malware data. Our feature importance analysis offered a good insight into identifying higher predictors of malware. Overall, our study provides a promising path toward improving malware detection capabilities, ultimately contributing to enhancement in computer security.