Improved Network Security in MANET using Dingo Optimizer with Attention Convolutional Neural Network-based Intrusion Detection and Classification Model

Abstract

The fast popularization and growth of Mobile Ad-Hoc Networks (MANET) raise numerous security concerns. An Intrusion Detection System (IDS) is an effective defense tool to identify malicious data in intricate network landscapes and ensure computer system security. Conversely, conventional IDS, based on classical ML models, lacks accuracy and reliability. Rather than using classical Machine Learning (ML) as in preceding research, Deep Learning (DL) can perform better in extracting features from large datasets and massive cyber traffic in the current context. In general, MANET has inferior physical security for mobile devices due to effects like a deficiency of centralized management, node mobility, and restricted bandwidth. To challenge these safety concerns, classical cryptography systems do not protect MANETs from new attacks and vulnerabilities, so employing DL models in IDS efficiently alters the potent environments of MANETs. It permits the method to decide on intrusion while continuing to study their mobile landscape. This study proposes an Improved Network Security using the Dingo Optimizer Algorithm-based Intrusion Detection and Classification (INSDOA-IDC) model in MANET. The main aim of the INSDOA-IDC technique is the effective detection and classification of intrusions in MANET. Initially, the INSDOA-IDC technique applies the Z-score normalization method for pre-processing the input data. Attention with a convolutional Neural Network-Bidirectional Long Short-Term Memory (A-CNN-BiLSTM) method is used for intrusion recognition and classification. Finally, the Dingo Optimizer Algorithm (DOA) is implemented to ensure the optimum selection of the hyperparameters connected to the A-CNN-BiLSTM model. Extensive simulations of the INSDOA-IDC method are accomplished under the NSLKDD and UNSW-NB15 datasets. The comparison study of the INSDOA-IDC model portrayed superior accuracy values of 99.53% and 99.55% under dual datasets over existing models.