Intrusion Detection based on extracting Optimization Features for Bidirectional Long-Short-Term-Memory

In today's insecure digital landscape, the need for robust Intrusion Detection Systems (IDS) is more critical than ever. Network attacks and unauthorized intrusions pose significant threats, requiring effective IDS development. Many intrusion detection algorithms combine traditional machine learning models with deep learning to enhance performance, but they still encounter issues like insufficient detection accuracy and potential accuracy reduction due to data preprocessing operations. This paper introduces a novel intrusion detection method that enhances Bidirectional Long-Short-Term-Memory (BiLSTM) performance using Logarithmic Autoencoder (LogAE) and Extreme Gradient Boosting (XGBoost) for optimal feature extraction, aiming for high detection rates. LogAE generates new features from input feature sets, which are then combined with the original features. XGBoost is employed to identify key features from this combined set, which serves as input to BiLSTM for robust attack classification. The proposed algorithms are evaluated on the NSL-KDD dataset using various evaluation metrics, demonstrating the effectiveness of the proposed method in improving IDS performance and providing valuable insights for secure and efficient intrusion detection systems against cyber security threats.