A Critical and Comprehensive Analysis of WSN Security Using Benchmark Datasets

WSN has become increasingly vital in several application scenarios over the last years, and thus assuring the security of these networks is a big concern. This paper introduces an integral comparative study of three of the most important machine learning algorithms: RF, SVM, and DNN, focused on intrusion detection in WSN environments. We measure these against four benchmark datasets: CICIDS2017, NSL-KDD, WSNDS, and UNSW-NB15. Our experimental results show that DNN constantly outperformed other algorithms on all datasets with the highest accuracies of 98.56% on CICIDS2017, 97.34% on WSNDS, 98.23% on UNSW-NB15, and 95.27% on NSL-KDD. Further, precision, recall, and F1-score metrics confirm the outstanding performance of DNN. Random Forest is the second-best performer while SVM shows relatively low but stable performance on all datasets. Our results show that deep learning-based approaches are quite robust and reliable security solutions for WSN environments when used in analyzing complex, up-to-date attack patterns. The insights gathered through this study will be a great source of information for researchers and practitioners in the implementation of effective intrusion detection systems for securing WSNs.