Optimized DDoS Detection in Software-Defined IIoT Using a Hybrid Deep Neural Network Model

Abstract

In the industrial internet of things (IIoT), DDoS attacks present a significant security challenge, requiring solutions that balance high detection accuracy with low computational cost. This study proposes a novel DDoS detection approach, IIoT Attack Detection based on CNN-mLSTM-KAN (IAD-CLK). By applying adaptive feature selection boosting (AFSB) during data preprocessing, the most relevant features are selected, reducing computational load. The CNN-mLSTM-KAN model combines depthwise separable convolutions, an mLSTM architecture enhanced with matrix operations, and the Kolmogorov–Arnold Network (KAN) to improve both detection performance and efficiency. Experimental results on the CICDDoS2019 dataset show an accuracy of 99.78% and a processing time of 0.122 ms, demonstrating the approach's effectiveness and suitability for IIoT environments.