Optimal Feature Set Analysis for Enhanced IoT Malware Detection with RFE and XGBoost

The swift integration of various industries through the Internet of Things (IoT) has escalated concerns about security, particularly regarding malware that has the potential to either exfiltrate data or interfere with operations. Numerous malware detection models have been developed; however, a significant concern remains the lack of emphasis on the speed at which threats can be detected. This paper introduces a malware detection model that combines choosing important features, balancing data, and machine learning. By examining the optimal number of features using a heuristic approach, the model can enhance detection speed without sacrificing the detection performance. Additionally, balancing techniques can address issues arising from uneven data distribution. The model classifies IoT traffic using XGBoost and is measured by six metrics: AUC, accuracy, precision, recall, F1-score, and execution time. Results show that using 11 features can classified data in 2.22 seconds, with 99.9952% AUC, 99.8106% precision, 99.8064% recall, 99.8064% F1-score, and 99.8048% accuracy.