Optimized Deep Ensemble Learning for DDoS Detection in IoT Networks via Pruning Techniques

Abstract

Distributed Denial of Service (DDoS) assaults pose a widespread risk to the safety and stability of IoT networks, regularly main to community congestion, downtime, and compromised facts integrity. This takes a look at proposes a unique deep ensemble learning approach for detecting DDoS assaults in IoT networks, incorporating pruning techniques to enhance model performance and efficiency. The ensemble model combines a Stacking Classifier, integrating all Machine Learning algorithms. Additionally, a TPOT Classifier is employed to automate and optimize the machine learning pipeline. The dataset used in this research includes network flow features such as flow duration, total forward and backward packets, packet lengths, flow inter-arrival times, packet statistics, and flag counts (e.g., SYN, FIN, ACK), providing comprehensive feature extraction for distinguishing between normal traffic and DDoS attack patterns. By utilizing pruning in the ensemble models, redundant and less significant models are eliminated, improving computational efficiency and predictive accuracy. The proposed machine is evaluated using performance metrics such as accuracy, precision, recollect, and F1-score, demonstrating its capability to accurately detect DDoS attacks while minimizing false positives. This technique no longer only complements the detection charge of DDoS assaults but also addresses the want for scalability and performance in resource-restricted IoT networks. Experimental outcomes indicate that the deep ensemble version, combined with pruning, achieves superior detection accuracy compared to traditional machine learning models, making it a robust solution for IoT network security.