Neural Computing and Applications Software

Gazelle-enhanced deep reinforcement learning for automated DDoS detection in SDN

Priyadharshini, M., Vanathi, B.
Springer Nature · 2026-11-14 · DOI 10.1007/s00521-026-12501-5

Abstract

Recently, software-defined networking (SDN) has gained much popularity and emerged as an emerging and promising networking paradigm. Unlike traditional network approaches, SDN offers enhanced security and resiliency towards multiple types of attack techniques. It mainly includes features such as unified network management, comprehensive network-wide knowledge and separation of control/data planes. But at the same time, there are many other threats that exist in SDN due to which SDN can be attacked. These threats include attacks on data plane, application plane, control plane, and transmissions within the SDN. In this paper, an efficient machine learning (ML) technique has been proposed to detect and predict different attacks in the proposed SDN model. Initially, the data obtained from public domain is normalized using the Tanh-Normalization (T-Norm) approach to eliminate any irrelevant null values. The proposed SDN model utilizes Improved Deep Reinforcement Learning (IDRL) with an adaptive Mountain gazelle optimizer (Adapt-MGO) for identifying the attacks. The proposed work eliminates false negatives by leveraging a fitness function. The framework was trained utilizing the NSL-KDD and CIC-DDoS2019 datasets. Various performance evaluation metrics have been used such as Accuracy, recall, Positive predictive value (PPV), False discovery rate (FDR) and Matthew’s correlation coefficient (MCC). The performance of the proposed algorithm has been evaluated, where 99.1%, 99.15% accuracy was achieved with 2.05 and 2.51 FDR for both datasets respectively.

Read at publisher →