Enhanced Intrusion Detection System through Optimizing Neural Wavelet Transforms and Pelican Optimization Algorithm-Long Short-Term Memory Techniques

Abstract

This research introduces an advanced intrusion detection framework employing a synergistic integration of preprocessing techniques and sophisticated machine learning models. The proposed methodology incorporates Neural Network (NN), Transverse Dyadic Wavelet Transform (TYDWT), and Fast Fourier Transform (FFT) as powerful preprocessing algorithms, refining and optimizing network datasets for subsequent analysis. Long Short-Term Memory (LSTM) networks, known for their sequence modelling capabilities, are employed for feature extraction and classification, showcasing adaptability to intricate patterns in network data. A significant contribution is proposed in the form of the Pelican Optimization Algorithm - Long Short-Term Memory (POA-LSTM), a novel algorithm designed for enhanced optimization and feature extraction. Notably, POA-LSTM demonstrates remarkable accuracy, marking a substantial advancement in intrusion detection capabilities. Additionally, the study explores the Cat Optimization Algorithm - Long Short-Term Memory (COA-LSTM), further extending the adaptability of LSTM models for intrusion detection. The holistic evaluation of the proposed framework encompasses essential metrics, including accuracy, precision, specificity and sensitivity, providing a comprehensive assessment of its performance across diverse intrusion scenarios. The results underscore the efficacy of the proposed POA-LSTM algorithm, emphasizing its capability to achieve high accuracy about 98% in intrusion detection.