Authors: Deepa Barethiya, Kajal Lanjewar, Damini Mondhe
Abstract: Malware — malicious software — represents one of the most pervasive and rapidly evolving threats in modern cybersecurity. Traditional signature-based detection systems, while effective against known threats, are fundamentally inadequate against polymorphic, metamorphic, and zero-day malware variants. This paper presents a comprehensive study and implementation of a machine-learning-based malware detection framework that overcomes the limitations of conventional approaches. The proposed system employs static analysis (PE header features, API call sequences, n-gram byte patterns), dynamic analysis (system call traces, network behaviors), and hybrid analysis to extract discriminative feature sets. Several supervised classification algorithms — including Random Forest, Support Vector Machine (SVM), Gradient Boosting (XGBoost), and a custom Convolutional Neural Network (CNN) — are evaluated on the EMBER 2018 and VirusShare benchmark datasets. Experimental results demonstrate that the ensemble model achieves a detection accuracy of 98.7%, a false-positive rate below 0.4%, and an average inference time of 12 ms, outperforming state-of-the-art baselines by a significant margin. The paper further discusses real-time deployment considerations, adversarial robustness, and future research directions.
