Article
EFFICIENT BOTNET ATTACK DETECTION IN IOT USING HYBRID MACHINE LEARNING MODEL
The rapid proliferation of Internet of Things (IoT) devices across various domains such as smart homes, healthcare, industrial automation, and smart cities has introduced unprecedented convenience but also exposed critical vulnerabilities to cyber threats. Among these threats, botnet attacks have emerged as one of the most severe and persistent challenges, leveraging large networks of compromised IoT devices to execute Distributed Denial of Service (DDoS), data theft, and other malicious activities. Due to the resource-constrained nature of many IoT devices, traditional security mechanisms are often insufficient or impractical for effective botnet detection. This project proposes a hybrid machine learning-based approach aimed at enhancing the detection accuracy and efficiency of botnet attacks in IoT environments. The hybrid model synergistically combines unsupervised learning techniques—which identify anomalies in network traffic without prior labeling—with supervised learning algorithms that classify traffic patterns into benign or malicious categories based on historical data. By leveraging the strengths of multiple algorithms, including Decision Trees, Random Forests, and Support Vector Machines, and integrating them through ensemble methods like stacking or voting, the model achieves robust detection capabilities against both known and novel botnet variants. The system architecture involves an initial data preprocessing stage that handles feature extraction, normalization, and selection from IoT network traffic datasets such as Bot-IoT and UNSW-NB15. The unsupervised component flags suspicious activities by clustering or anomaly scoring, while the supervised component refines detection by learning from labeled examples. This layered detection approach reduces false positives and enhances detection speed, which are critical for real-time IoT security applications. Experimental results demonstrate that the hybrid model significantly outperforms individual machine learning classifiers in terms of accuracy, precision, recall, and F1- score. Furthermore, the model maintains low detection latency, making it suitable for deployment in real-world IoT environments where timely responses to attacks are essential.
Full Text Attachment
