Data-centric explainable artificial intelligence techniques for cyber-attack detection in microgrid networks

Abstract

This article presents a strategy for pre-processing data and interpreting the performance of advanced machine learning (ML) models to improve data quality and modify the models based on identified problems during the iterative development of classification models. The process consists of three main stages aimed at maximizing the performance of ML models. Stage-1 involves data collection and preprocessing. Stage-2 focuses on improving data quality and extracting features. The extracted features are further refined using hyperparameters in Stage-3. Additionally, the framework utilizes SHapley Additive exPlanations (SHAP) for explainable artificial intelligence (XAI) to elucidate the learning process of ML models and, therefore, make the model's behavior interpretable. As an example, this innovative process has been applied to cyber-attack detection in microgrid networks. In Stages-1, a CIGRE low-voltage microgrid network is simulated for data collection during cyber-attacks, followed by data pre-processing. In Stage-2, data augmentation is performed using SMOTE and ENN, and feature extraction is carried out using the Boruta Python package. In Stage-3, hyperparameters are tuned using the TPE algorithm. The results demonstrate the effectiveness of this approach, enabling the model to progressively enhance its predictive capacity through each stage.