Learning to Protect Machine Learning-Based Intrusion Detection Systems for Enhanced Security in MANETs
Main Article Content
Abstract
Mobile Ad Hoc Networks (MANETs) are dynamic and decentralized, making it difficult to implement robust and adaptable Intrusion Detection Systems (IDS). This study examines the effectiveness of Support Vector Machines (SVM), Decision Trees, Random Forests, K-Means Clustering, Autoencoders, and a Proposed Technique in MANET security. An in-depth study includes Detection Accuracy, Precision, Recall, F1 Score, AUC-ROC, and Specificity. Detection Accuracy is 95%, precision, recall, and F1 Score are excellent with the Proposed Technique. It is resilient to network fluctuations and adversarial attacks, making it an attractive real-world deployment option. Decision Trees and K-Means Clustering are efficient computational choices for resource-constrained MANETs. Hybrid models with supervised and unsupervised learning improve IDS flexibility. For changing MANET attack scenarios, labeled and unlabeled data can improve detection accuracy. Interpretability remains difficult, especially for sophisticated models like Autoencoders, despite these advances. MANET-specific interpretable ML models should be the focus of future research.