Machine Learning Approaches for Detecting IoT Botnet Attacks: A Comparative Study of N-BaIoT Dataset

This study compares traditional machine learning models using the N-BaIoT dataset to identify botnet attacks on the Internet of Things. Regardless of not being among the IoT security measures, this research highlights the importance of effective detection mechanisms by using six machine learning models (SVM) Support Vector Machine, (LR) Logistic Regression, (KNN) K-Nearest Neighbors, (NB) Naive Bayes, (DT) Decision Trees, and (RF) Random Forest which were evaluated on 09 (nine) IoT devices network traffic datasets specifically Bashlite, and Mirai malware. In this comparative study among all the models, the Random Forest consistently demonstrated prominent performance, Reaching 99% in overall accuracy, the model also demonstrated a precision of 0.90, a recall of 0.79, and an F1-score of 0.79, reflecting strong predictive capabilities on 9 IoT devices, including security and surveillance cameras. In contrast, SVM and Linear Regression struggled with nonlinear traffic patterns, with SVM acquiring lower accuracy (54%) on the security camera. The study concludes with further research recommendation for model refinement of the real-time application, and malware dataset expansion to enhance the overall adaptability and detection accuracy of these machine learning models.