Attack Detection on Edge IIOT Using Ensemble Learning and XAI Model Transparency

The Industrial Internet of Things (IIoT) has rapidly developed, improving production efficiency and profitability and introducing significant security challenges as IIoT systems become more vulnerable to cyberattacks. This study addresses these challenges by implementing attack detection using ensemble learning methods and Explainable Artificial Intelligence (XAI). Models like Random Forest, XG- Boost and LightGBM are employed with SHAP (SHapley Additive Explanations) to improve model transparency. Stratified Sampling reduces data volume while preserving feature distribution, and feature selection is performed using Random Forest Feature Importance to achieve high accuracy and efficiency. The Edge-IIoTset dataset is used, with preprocessing and feature selection optimized through Random Forest Feature Importance. Testing involves two scenarios with varying feature counts, and data splits to evaluate effectiveness. Results show that the LightGBM model achieved the highest accuracy 97.60%, followed by XG-Boost 96.90% and Random Forest 96.51%. Additionally, SHAP identified key features influencing predictions and improving user trust and understanding. These results demonstrate how well ensemble learning and XAI work together to improve IIoT security.