A Robust Malware Classification Approach Leveraging Explainable AI

Malware detection is a critical aspect of cybersecurity, with the constant evolution of malicious software posing significant challenges to computer systems and networks. Traditional security measures often struggle to detect obfuscated malware, which employs sophisticated techniques to evade detection. This research explores the utilization of memory analysis techniques coupled with machine learning and deep learning algorithms, augmented by Explainable AI (XAI), to enhance malware detection and classification. Leveraging a well-balanced dataset specifically tailored for memory analysis, various algorithms including Random Forest, XGBoost, Logistic Regression, LSTM, GRU, ANN, and hybrid models were employed for multiclass classification. Ensemble methods were also implemented to improve classification accuracy. The results demonstrate the efficacy of ensemble methods like Random Forest and XGBoost in accurately detecting obfuscated malware, with Random Forest achieving an impressive accuracy of 87.30%. Furthermore, the integration of GRU and LSTM architectures in hybrid models showed promising results, highlighting the effectiveness of combining different neural network architectures for improved performance in multiclass classification tasks. Through Explainable AI techniques such as LIME, insights into the underlying features and decision criteria influencing classification outcomes were provided, enhancing the interpretability and transparency of the models. This research contributes to advancing the field of malware detection and combating evolving cyber threats, with practical implications for real-world cybersecurity applications.