Performance Benchmarking of ANN and LSTM Architectures for Network Intrusion Detection Using Chi-Square Feature Optimization

Abstract

The increasing complexity and frequency of cyber-attacks continue to pose significant challenges to network security, thereby necessitating the development of efficient and reliable Intrusion Detection Systems (IDS). To address this challenge, this study proposes deep learning-based Network Intrusion Detection Models (NIDMs) using Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM), integrated with Chi-Square feature selection for improved model efficiency. The study adopts a structured three-phase approach comprising data acquisition and preprocessing, model implementation, and performance evaluation. To ensure robustness and enhance the generalizability of the proposed models, the evaluation is conducted on two distinct benchmark datasets, NSL-KDD and WUSTL, representing different network environments and attack characteristics. This dual-dataset validation framework enables a more comprehensive assessment of model performance beyond single-dataset limitations. Experimental results demonstrate that the ANN-based model achieved accuracies of 99.99% and 99.96% on the NSL-KDD and WUSTL datasets respectively, while the LSTM-based model achieved 99.95% and 96.5% on the same datasets. These findings indicate that the proposed approach not only improves detection accuracy but also maintains consistent performance across heterogeneous datasets, thereby confirming its effectiveness and practical applicability for real-world intrusion detection scenarios.