An Adaptive and Explainable Artificial Intelligence Framework for Intelligent Computing, Predictive Analytics, and Autonomous Decision-Making in Next-Generation Digital Systems
Abstract
This study developed an adaptive and explainable artificial intelligence framework for intelligent computing, predictive analytics and autonomous decision-making in next-generation digital systems. The framework was tested with a PV predictive-maintenance data set of 500 observations that consist of 14 variables measuring the physical state of the PV, the health of the inverter, the severity of the fault, the constraints of the maintenance, the quality of the data, the priority of the decision, and the actions taken in the maintenance. Stratified sampling was used to train five supervised learning algorithms: Logistic Regression, Decision Tree, Random Forest, XGBoost and LightGBM, and SHAP analysis was used to obtain global and local model explanations. The results indicated that XGBoost outperformed the other classifiers in classification accuracy at 97.0%, while Random Forest had the best overall balance, with an F1 score of 0.885, a balanced accuracy of 0.965, and an ROC-AUC of 0.993. The soiling index, fault severity, inverter health, and inverter status were the most significant features influencing maintenance decisions, according to SHAP analysis. The following frequency of urgency for maintenance actions was also shown in the decision-priority analysis: Monitor > Inspect > Clean and Repair. In general, the proposed framework has successfully integrated predictive modelling, explainability, and automated decision recommendations, and offers significant practical value for implementing transparent and adaptive maintenance support in intelligent digital energy systems.