Enhancing IoT Botnet Detection: A Comparative Study of Balancing Techniques with Naive Bayes Classifier
Hartono Saputro, Tohari Ahmad, Muhammad Aidiel Rachman Putra · 2024
The Internet of Things (IoT) revolution has increased the need for robust cybersecurity measures due to the increased vulnerability of interconnected devices to botnet attacks. Attack datasets are always imbalanced, which affects detection performance. This research investigates the effectiveness of Naive Bayes classifiers in detecting IoT botnets, focusing on the impact of SMOTE, ADASYN, and Borderline-SMOTE ensemble balancing techniques on classification performance. Using the IoT-23 dataset, this study systematically evaluates these techniques in terms of precision, recall, F1-score, and overall accuracy. Despite these challenges, SMOTE emerged as a relatively better balancing technique, with a high precision in class identification of 99.19%. The recall of 72.45% confirms its ability to recognize positive class instances, the F1-score of 79.13% shows an optimal balance between precision and recall, and the accuracy of 72.45% shows consistency. This study contributes valuable insights into enhancing cybersecurity for IoT networks through improved botnet detection methodologies,