A Preemptive Ensemble Learning Framework for KnowledgeBased Network Intrusion Detection

Abstract

  The rapid growth of Internet-connected systems has increased the complexity, diversity, and volume of network traffic, making intrusion detection a significant cybersecurity challenge. Traditional machine learning-based intrusion detection systems often suffer from high-dimensional data, severe class imbalance, and poor detection of low-frequency attacks. To address these limitations, this paper proposes a hybrid framework, FD-ANN, which integrates Density Peak Clustering (DPC), Fractal Fuzzy Membership (FFM), and Artificial Neural Networks (ANNs). The framework employs wrapper-based feature selection to eliminate irrelevant features and reduce dimensionality. DPC partitions the training data into homogeneous clusters, preserving minority attack structures while mitigating class imbalance. Separate ANN classifiers are trained for each cluster, and their outputs are combined through an adaptive FFM weighting mechanism that dynamically assigns weights according to local data distributions.

Experiments on the NSL-KDD and UNSW-NB15 benchmark datasets demonstrate the effectiveness of the proposed approach. FD-ANN achieved accuracies of 82.47%, 92.24%, and 95.59% on NSL-KDDTest-21, NSL-KDDTest+, and UNSW-NB15, respectively. The framework also significantly improved the detection of minority attack classes while reducing false alarm rates. Comparative results against KNN, RF, SVM, DT, and conventional ANN models confirm that FD-ANN provides an effective, scalable, and robust solution for modern network intrusion detection.