Feature selection based SOOA-SDRGRU for DDoS attack detection and classification

Abstract

Distributed denial-of-service (DDoS) attacks monitor network traffic to determine unusual patterns that represent malicious flooding. However, DDoS attacks mimic legitimate traffic which makes it difficult to accurately detect and classify between normal and attack. This research proposes Somersault Orangutan optimization algorithm for feature selection with ShakeDrop refined gated recurrent unit (SOOA-SDRGRU) to detect and classify DDoS. In OOA, somersault foraging behavior is incorporated to solve local optima issue which enhances solution diversity and rapid convergence speed toward global optima. GRU captures temporal patterns and dependencies in sequential network data, whereas ShakeDrop assist to minimize overfitting, which leads to more accurate and reliable detection of diverse attack patterns. Compared to existing methods like bacterial colony optimization (BCO) with optimized back propagation neural network (BPNN), the proposed SOOA-SDRGRU obtains high accuracy of 0.9989 and 0.9992 on CIC-IDS2017 and CIC-DDoS2019 datasets which shows robust detection method for evolving DDoS patterns.