DDoS Detection Using Artificial Neural Network Regarding Variation of Training Function

Distributed denial-of-service (DDoS) is a type of network attack with the number both in volume and intensity has increased significantly in recent years. DDoS is a major problem for the integrity, secrecy and availability of resources owned by Internet organizations. Early detection of DDoS attacks is a fundamental process performed automatically by the Intrusion Detection System (IDS), which generally uses signature-based detection techniques that can be said to be far from perfect when compared with the more modern cyber attack techniques. This study proposed DDoS detection system based on network feature which produced from statistical extraction combined with artificial neural network (ANN) method as detection engine with training function variation. The experiment resulted that Quasi-Newton training function (Matlab trainlm) give the highest accuracy value 0.992 (99.2%) against Resilient-Propagation training function (Matlab trainrp) which resulted accuracy at 0.989 (98,9%) and the Scaled-Conjugate training function (Matlab trainscg) which resulted accuracy at 0.988 (98,8%).