Boosting-based Models with Tree-structured Parzen Estimator Optimization to Detect Intrusion Attacks on Smart Grid

Smart grid is an emerging technology that transfers power to users intelligently through two-way communication. Despite the benefits of this network, it is prone to different cyber-attacks. One solution to address this issue is the use of intrusion detection systems. Several studies have been conducted to investigate the shortcomings of such system, which include low detection rates and high false alarms; however, these studies did not completely address these issues. Motivated by the existing gaps, we investigate the performance of boosting-based models, namely Adaptive Boosting, Gradient Boosting, and Categorical Boosting, in detecting cyber-attacks on smart grid networks. The performance evaluation is conducted based on accuracy, probability of detection, probability of misdetection, and probability of false alarm. The results of the models were compared with those of three widely used traditional machine learning models, namely support vector machine, naïve Bayes, and K nearest neighbor. The benchmark of CICDDoS 2019 is selected as a dataset for training, validation, and testing. The ReliefF feature selection technique is used to identify the most important features for training the models. We also used the Tree-structured Parzen Estimator optimization technique to find the best hyperparameters for each model and ensure optimal performance. The results show that the boosting-based models outperform the three traditional models, and the Categorical Boosting classifier has the best results in terms of the four-evaluation metrics.