A Unified Multi-Model Learning Framework for Reliable Static Malware Detection

Malware has emerged as a key threat to computer systems and networks, and it is essential to achieve accurate and reliable malware detection. In this article, a unified multi-model learning framework is introduced that integrates deep learning and classical machine learning techniques to provide a comprehensive approach to detecting static malware. Experiments were conducted on a high-dimensional dataset (799912 rows, 2382 columns, 50000 rows) of static malware features using multiple models that include deep neural network models such as MLP, MalConv-X, CNN Hybrid, and classical models such as Logistic Regression, Random Forest, and LightGBM. Each model is trained and evaluated using evaluation metrics such as accuracy, precision, recall, f1-score, and AUC to ensure fair comparison and assessment. The results show that Light GBM achieved the highest performance with an accuracy of 95.48% and an AUC of 0.9915. Thus, LightGBM achieved the highest discriminative performance between malware and benign files. Deep learning models such as MLP and MalConv-X also performed well, showing 0.92 f1-score after training over 10 epochs. The The CNN-hybrid model showed the highest precision value of 0.9459 but a comparatively lower recall value of 0.8721. Correlation metrics, radar charts, and epoch-wise results indicate that ensemble learning models achieve strong performance in multiple evaluation parameters, and on the other hand, deep learning models exhibit stable convergence behavior during training. The proposed unified multi-model framework shows a reliable performance for static malware detection and provides a practical approach for model selection in real-world cybersecurity applications.