Parallel ANN Model: A Potential Alternative to CNNs for Network Intrusion Detection on the UNSW-NB15 Dataset

Intrusion Detection Systems (IDS) are essential for defending networks against advanced cyber threats. Traditional machine learning and deep learning models like Convolutional Neural Networks (CNNs) often underperform on highly imbalanced datasets, especially for rare attacks. We introduce a Parallel Artificial Neural Network (P-ANN) architecture using the UNSW-NB15 dataset. P-ANN employs multiple parallel branches that independently process feature subsets before merging into a unified classification layer, enhancing feature representation and generalization. Class weighting is applied during training to address imbalance in both binary and multiclass tasks. Model interpretability is provided by Local Interpretable Model-Agnostic Explanations (LIME), allowing insights into feature importance and decision logic. P-ANN achieved test accuracies of 91.46% (binary) and 79.23% (multiclass), outperforming baselines by 2.55% and 4.07%, respectively. It also improved precision, recall and F1 scores across attack categories while reducing training time and parameter count. These results demonstrate P-ANN’s potential as an efficient, interpretable solution for practical intrusion detection.