A Comparative Study on Malware Detection Using Supervised Machine Learning Models

Traditional signature-based systems struggle to detect novel and variably structures threats such as polymorphic and metamorphic malware. These systems rely on predefined rules, which limit their ability to identify newly developed, obfuscated, or zero-day attacks. Given the constantly evolving nature of cyber threats, it is crucial to develop detection systems capable of identifying malicious behavior without relying solely on static signatures. This study investigates the effectiveness of supervised machine learning (ML) techniques in detecting malware using the CICIDS2017 dataset which includes both attacks and benign traffic. Four widely used supervised models, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors(KNN) and XGBoost, are evaluated and compared. Each model undergoes the same data preparation process, including features selection and data balancing, to ensure fair performance assessment. Model Performance is evaluated using standard metrics such as accuracy, precision, recall and F1-score. Among the models, Random Forest achieved the highest accuracy of approximately 99.8%, demonstrating strong robustness and generalizability. XGBoost followed with a commendable accuracy of around 92%, offering a balance between computational efficiency and interpretability. In contrast, SVM and KNN exhibited limitations in detecting minority attack classes. Overall, the Random Forest model outperformed other established
methods. methods. Feature importance analysis revealed that
attributes such as Avg Bwd Segment Size and Flow IAT Max
significantly contribute to the detection of malicious traffic.