MULTI-MODEL HYBRID MACHINE LEARNING APPROACH FOR IOT BOTNET ATTACK DETECTION

Abstract

Internet of Things botnets are capable of launching DDoS attacks, breaching networks, and stealing data. Traditional detection methods and single-model machine learning are unable to detect intricate attack patterns due to the high-dimensionality, imbalance, and dynamic nature of network traffic data. This investigation enhances the precision and robustness of a multi-model hybrid machine learning IoT botnet detection system through the application of deep and supervised learning. Stacking and weighted voting are employed by SVM, LSM, and Random Forest to monitor temporal and geographical traffic. Performance and noise are enhanced through the selection of features and the subsequent preprocessing. Rival algorithms were surpassed by the model in the areas of precision, recall, accuracy, and false positives in large IoT datasets. A continuous detection of IoT botnets is achieved through a flexible hybrid architecture.