Advanced Network Security: An Enhanced BI-LSTM Model for Intelligent Intrusion Detection

Abstract

Intrusion Detection Systems (IDS) play a crucial role in safeguarding network security against evolving cyber threats. Traditional IDS models often suffer from high false alarm rates and inefficient detection, necessitating the adoption of advanced deep learning techniques. This paper presents an Enhanced Bidirectional Long Short-Term Memory (Enhanced BI-LSTM) model for intrusion detection, integrating Feature Selection, Attention Mechanism, and Regularization to improve accuracy and reduce computational overhead. The proposed model utilizes Principal Component Analysis (PCA) and Chi-Square tests for optimal feature selection, while the Attention Mechanism enhances learning by focusing on critical time steps. The KDD Cup 1999 dataset is used for training and evaluation, containing diverse intrusion types. Experimental results demonstrate that the Enhanced BI-LSTM achieves an accuracy of 98.5%, outperforming conventional models such as SVM, QDA, and standard LSTMs. The model also achieves a lower Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), ensuring robust detection performance. This study highlights the effectiveness of deep learning-based IDS in real-world cybersecurity applications.