Deep stacked autoencoder with fractional VCROA for DDoS attack detection using a big data approach in the MapReduce framework

The rising dependence on internet-based services has exposed network infrastructure to increased vulnerability to cyberattacks, especially DDoS attacks. The attacks flood target systems with unwarranted traffic that disrupts legitimate access and undermines service reliability. To overcome this issue, the present paper proposes an optimization-based deep learning model, called Fractional Velocity Contour-based Remora Optimization Algorithm-Deep Stacked Autoencoder (FVCROA_DSA), for high-efficiency DDoS attack detection in a MapReduce environment. The model combines a mean-substitution method for filling data gaps and Support Vector Machine Recursive Feature Elimination (SVM-RFE) in the mapper step to identify the most significant network attributes. This step is followed by the reducer stage, which trains a Deep Stacked AutoEncoder (DSA) to recognize attack patterns, which is then fine-tuned by the proposed FVCROA algorithm. Fractional Calculus leads to increased optimization stability and faster convergence during training. Experimental tests on the BOT-IoT and DDoS Attack datasets show that the FVCROA architecture with DSA achieves higher detection accuracy, with a precision of 93.857, a recall of 94.827, and an F-measure of 94.340, surpassing the current baseline techniques in scalability and reliability.