Optimizing WSN Security: An Advanced ML-Driven Intrusion Detection System with SMOTE-Tomek

This paper presents a novel intrusion detection framework for Wireless Sensor Networks (WSNs), which incorporates advanced Machine Learning (ML) techniques along with the SMOTE-Tomek algorithm for synthetic minority oversampling and noise reduction. Our approach addresses the critical challenges of dataset imbalance in WSN security. The framework employs a comprehensive suite of ML algorithms, including decision trees, random forests, support vector machines (SVM), k-nearest neighbors (KNN), and ensemble methods, alongside robust data processing strategies. Key components of our methodology include SMOTE-Tomek for dataset balancing, enhancing representation of both normal and anomalous instances, Advanced data preparation techniques, such as feature standardization, data cleaning, and compression, and Sequential Backward Selection (SBS) for optimal feature selection, improving computational efficiency. We evaluated our model using the WSN-DS dataset, containing 374,661 records. The results demonstrate exceptional performance: Quadratic SVM achieved near-ideal accuracy in both binary and multiclass classification tasks, our approach significantly outperformed conventional algorithms in terms of detection precision and false alarm reduction, and Comprehensive performance metrics, including accuracy, precision, recall, and F1 score, Assessing the model’s performance. This proposed work introduces a meaningful improvement in intrusion detection mechanisms for wireless sensor networks (WSNs) , offering enhanced security, adaptability, and operational resilience for critical infrastructure applications. The proposed framework achieved 100% accuracy on the WSN-DS dataset and 97.3% accuracy on the UNSW-NB15 dataset, with a false positive rate below 1.2%. By effectively addressing dataset imbalance and leveraging state-of-the-art ML techniques, our approach paves the way for more robust and efficient security solutions in WSNs.