Optimizing Intrusion Detection in Software-Defined Networks Through Automated Machine Learning and Intelligent Feature Engineering
Dhananjay Bisen, Anshul Ghanghoria, Praneet Saurabh, D. Rohith, Upendra Singh · IEEE Access · 2025
Software-defined networking is a network management approach that allows flexible and efficient network design through programming. It adds flexibility to network design for increasing performance and monitoring. This flexible design of software-defined networking presents unique security challenges, especially in detecting and mitigating intrusions. Traditional intrusion detection systems are often manual and rule-based, struggling to adapt to the dynamic nature of software-defined networking environments. Recent years have witnessed the use of various machine learning approaches for intrusion detection in software-defined networking, but it also requires manual expertise. Automated machine learning (AutoML) is a compelling approach that automates the process of finding the best architecture to find the anomalies in software-defined networking and overcomes the requirement of manual expertise. This research paper proposes an innovative approach using AutoML with Auto feature engineering techniques (AutoFE) to enhance the effectiveness and efficiency of intrusion detection in software-defined networking. AutoML with a combination of AutoFE, hyperparameter optimization, and model selection results in substantial improvements in accuracy and efficiency of the model. Additionally, a ten-fold cross-validation was applied, alongside benchmarking against alternative AutoML platforms, namely Auto-sklearn, TPOT, H2O, and FEDOT, executed under identical parameter settings. Experimental results show that the optimized model using a Decision Tree accomplished an outstanding accuracy of 99.976% and precision of 99.987% on the benchmark dataset, outperforming the other conventional models. These results reveal the impact of AutoML with AutoFE in building efficient and accurate models for software-defined networking intrusion detection. These findings highlight how AutoML, when combined with AutoFE, accelerates model-building for intrusion detection in software-defined networking environments.