An Optimized LightGBM Framework for Static Malware Classification Based on Windows Portable Executable File Attributes

Malware poses a significant threat to the digital world, as weak detection systems can enable attackers to steal data or corrupt critical files. Numerous researchers have contributed to this domain by developing highly accurate malware classification systems using various machine learning and deep learning techniques. According to the literature, most machine learning models have been extensively explored for accurate malware classification, with LightGBM consistently demonstrating superior performance in most cases. However, there remains scope for developing more efficient, lightweight, and robust malware classification models through machine-learning-based ensemble techniques. In this paper, we conduct experiments using the balanced and extensive EMBER 2018 dataset, comprising 799,876 samples with 2,382 features, to ensure robust training and obtain an unbiased model. We fine-tune the LightGBM model and additionally implement various machine learning models including Random Forest (RF), ExtraTrees Classifier (ET), XGBoost Classifier (XGB), and a soft-voting ensemble stacking RF, ET, and LightGBM classifiers. Our results show that the proposed optimized fine-tuned LightGBM model outperforms other approaches, achieving an accuracy of 96%.