An Optimization of The Hyperparameters of Multi-Valued Neutrosophic ConvLSTM for Intrusion Detection

In today’s world, cybersecurity has become indispensable as networks play an integral role in daily life. The increasing size and complexity ‎of these networks heighten the risk of new types of attacks, making it crucial to design robust and efficient Intrusion Detection Systems ‎‎(IDS). Numerous Deep Learning (DL) techniques have been developed to automate IDS and detect abnormal network behaviors. But certain models might be inaccurate during the training phase due to the presence of numerous irrelevant features. To solve these problems, ‎Feature Multi-Valued Neutrosophic Convolutional Long Short-Term Memory (FMVN-ConvLSTM) was created. But the hyperparameters ‎of ConvLSTM were not optimized, which limits its intrusion accuracy and increases model complexity. This paper leverages the Dove ‎Swarm Optimization (DSO) model to fine-tune ConvLSTM hyperparameters for enhancing IDS accuracy and reducing computational ‎complexity. Inspired by the flocking behavior of doves, DSO dynamically adjusts key ConvLSTM parameters like rate of learning, size of ‎batch, weight decay, dropout rate, and optimizer to achieve an ideal balance between performance and efficiency. DSO navigates the hy-‎hyperparameter space by mapping initial dove positions to starting configurations, refining them iteratively to facilitate efficient model learning. ‎Unlike other optimization strategies, DSO enhances population diversity and streamlines model complexity, allowing ConvLSTM to capture ‎essential temporal and spatial patterns. This approach produces a robust IDS capable of identifying various network threats with high accuracy and reduced computational overhead. The complete model is termed an Optimized FMVN-ConvLSTM (OFMVN-ConvLSTM). In the ‎final analysis, the experiment results show that compared to other current models on three distinct datasets, CIC-IDS2018, WSN-DS, and ‎UNSW-NB15, the OFMVN-ConvLSTM achieves an accuracy of 97.48%, 96.35%, and 96.85%‎.