A supervised machine learning-based solution for efficient network intrusion detection using ensemble learning based on hyperparameter optimization

An efficient machine learning (ML) ensemble technique for categorizing Intrusion Detection (ID) is proposed in this study. The tuning of the ML model’s parameters is a critical topic since it can improve detection quality. Another area where quality might be enhanced is pre-processing. Corrections to the training dataset can help with class identification, especially for unusual attacks like R2L (Root to Local attacks), U2R (User to Root attack). When compared to existing methodologies, the proposed methodology has a number of advantages, such as (1) it proposes two methods for classifying intrusions on the two most widely used datasets using ML models. (2) The KDD Cup99 and NSL-KDD datasets are rebalanced through data augmentation. (3) Provides a 3 steps approach for improving detection of intrusion utilizing Multi Layer Perceptron (MLP) in a cascaded structure. (4) To classify each class using a specialized one, a cascaded meta-specialized classifier architecture has been developed. (5) All meta-specialists assess the dataset’s non-flagged connections. With a classification accuracy of 89.32% and an FPR of 1.95%, this approach has been shown to considerably increase detection quality. (6) Finally, to enhance detection capability, the best algorithms’ predictions are integrated by increasing their weights. On the NSL-KDD dataset, this approach has a high accuracy of 87.63% and a low FPR of 1.68%. © 2022, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.