Energy Valley Optimized Feature Selection with CatBoost Classifier for Intelligent Industrial IoT Intrusion Detection
Abstract
IIoT systems inherently susceptible to assaults, their widespread use has increased pressure for sophisticated cybersecurity solutions. An effective classifier, innovative feature selection method, and state-of-the-art data preparation are all part of intrusion detection framework that is presented in this paper. Preprocessing WUSTL-IIoT dataset involves label encoding and normalization to make it compatible with machine learning techniques. dataset contains complicated data from industrial networks. Energy valley optimizer (EVO), a recently suggested metaheuristic that draws inspiration from gravitational potential energy landscapes, is used for feature selection to decrease redundancy and improve classifier performance. Reducing computing load and improving detection accuracy are goals of this technique, which seeks to keep only most relevant characteristics. Because of its reputation for handling categorical features and preventing overfitting, CatBoost algorithm is used to classify optimized features. suggested EVO-CatBoost pipeline outperforms other models compared to it, including random forest, support vector machine (SVM), and XGBoost, when joint through classic optimisers such as PSO, GWO, and ACO. As compared to all baselines, model's 95.47% accuracy, 95.00% F1-score, and 0.961 AUC-ROC were phenomenal. effectiveness of EVO in improving CatBoost's feature selection and classification capabilities for IIoT contexts is validated by these results. Results indicate that intrusion detection scheme may be easily scaled and is effective enough for use in real-time industrial settings