Intelligent Intrusion Detection in Software-Defined Networks Using a Hybrid Deep Learning Model with Feature Selection and Adaptive Thresholding
Abstract
Software-Defined Networks (SDNs) are gaining widespread adoption due to their centralized management and flexibility. However, these same traits make them vulnerable to sophisticated security threats. Traditional intrusion detection systems (IDSs) often fail to cope with the dynamic nature of SDNs and generate high false positive rates. This paper introduces an intelligent IDS framework combining hybrid deep learning (CNN + LSTM), feature selection, and adaptive thresholding. Feature selection reduces data dimensionality for improved efficiency, while adaptive thresholding dynamically adjusts decision boundaries to minimize false alarms. Evaluations on NSL-KDD and CICIDS2017 datasets show that the proposed system outperforms standalone models, offering superior accuracy and lower false positives.