An Efficient Intrusion Detection Model with Novel Machine Learning Stacking Ensemble Techniques

Cybersecurity threats are becoming increasingly sophisticated, necessitating the development of robust intrusion detection systems. Traditional methods for managing intrusion detection may not consistently yield satisfactory results. The adoption of ML based methods in intrusion detection is crucial due to the limitations of conventional techniques. This paper proposes the development of ML based Intrusion Detection model that seamlessly integrates ML ensemble techniques to enhance efficiency in identifying cyber threats. A dataset from Kaggle was used for experiments. Initially, Several ML classifiers namely K-Nearest Neighbors, Naïve Bayes, Support Vector Machine, Decision Tree and Random Forest are applied for Intrusion Detection and achieved best accuracy of 91% with Random Forest. To increase the accuracy, ensemble learning applied with combination of several ML algorithms and achieved increased accuracy when compared with single ML model. Two ensemble models namely Cost-sensitive Stacking and Ensemble Distillation are proposed and achieved accuracies of 94%, 96% respectively. The experiments shown that the proposed ensemble methods outperform conventional approaches for Intrusion Detection.