Effective Intrusion Detection in WSNs: Evaluating Machine Learning Models on WSN DS and WUSTL EHMS-2020 Datasets

This study addresses the detection of DoS (Denial of Service) attacks in WSNs (Wireless Sensor Networks), which are constrained by limited resources and node vulnerabilities. The primary goal is to improve the precision and reliability of IDSs through advanced machine learning techniques. Firstly, data preprocessing involved handling missing values, applying feature scaling (standardization and min-max normalization), encoding labels, and balancing the dataset using the SMOTE (Synthetic Minority Over-sampling Technique). Secondly, feature selection was conducted using Recursive Feature Elimination (RFE) to ensure high-quality input data. The study applies a range of machine learning algorithms, including MLP Classifier, Decision Tree, K-Nearest Neighbors, Gaussian Naive Bayes, SGD Classifier, Logistic Regression, Random Forest, and Voting Classifier, to assess their performance for the classification task. WSN-DS and WUSTL-EHMS-2020 datasets are used to evaluate effective intrusion detection systems (IDSs) for WSNs. Results showed that the Ensembles Learning methods achieved the highest accuracy, while tree-based models were particularly effective in detecting DoS attacks. These findings highlight the need for further research on real-time and hybrid detection strategies to enhance IDS performance in WSNs.