Detecting DoS Attacks in Sensor Networks Using Deep Learning for Improved Cybersecurity

Abstract

Wireless Sensor Networks (WSNs) constitute an attractive target as they are increasingly being incorporated into Internet of Things (IoT) applications, where they are vulnerable to the types of security threats, such as Denial of Service (DoS) attacks. These attacks are hard to detect to ensure the dependability and functionality of the WSNs. This paper suggests an improved machine learning (ML) and deep learning (DL) solution to detect DoS through the WSN-DS dataset. The methodology contains data preprocessing, feature and target definition, label encoding, dimension reduction through Principal Component Analysis (PCA), data balancing using SMOTE and train-test splitting. The ML models, including the Logistic Regression, Random Forest, Gradient Boosting, Decision Trees, Extra Trees, Naive Bayes, K-Nearest Neighbors, AdaBoost, LightGBM, XGBoost, and CatBoost are hyperparameter-optimized with Ant Colony Optimization (ACO). MLP, CNN, RNN, LSTM, GRU models are also tuned with ACO on the number of epochs and the batch size. Analysis in terms of accuracy, precision, recall, F1-score, AUC, specificity, and MCC reveals that the optimized approach performs much better in identifying DoSs and can therefore be adapted to increase the security of WSNs and enable IoT intrusion detection studies.