Feature Selection in Intrusion Detection System using Levy Flight based Crested Porcupine Optimization

Intrusion Detection Systems (IDS) are crucial for protecting interconnected networks by detecting malicious attacks and activities in Internet of Things (IoT). Nevertheless, as the amount of information increases, dimensionality reduction becomes critical for enhancing the efficiency of Machine Learning (ML) approaches during training. Therefore, this manuscript proposes the Levy Flight-based Crested Porcupine Optimization (LF-CPO) approach for feature selection in IDS. This research firstly gathers data from multiple datasets, including UNSW-NB15, Bot-IoT, CICIDS2017, and NSL-KDD, to validate the effectiveness of the introduced approach. Pre-processing approaches such as one-hot encoding and Min-max normalization are applied to enhance model performance. The Refined Long Short-Term Memory (RLSTM) approach is then employed for classifying intrusion attacks from the selected features. The experimental results demonstrate that the proposed LF-CPO with RLSTM approach achieves a superior accuracy of 98.89% and precision of 97.72% on the UNSW-NB15 dataset, outperforming the existing Extreme Gradient Boosting (XGB) technique.