An Integrated AI, IOT, and Digital Twins Framework for Engineering Optimization and Strategic Business Intelligence in the Philippines

Abstract

This study is designed and assessed an Artificial Intelligence (AI) supported Internet of Things (IoT) and Digital Twin approach for Predictive Business Intelligence, following Design Science Research (DSR). The conceptual artefact combines IoT data acquisition, Digital Twin state representation, machine-learning analytics and a Business Intelligence decision-support layer. The UCI AI4I 2020 Predictive Maintenance Dataset with 10,000 observations is used in empirical evaluation. Precision, recall, F1 score, PR AUC, ROC AUC, calibration, and confusion matrix are used to compare performance of Logistic Regression, Decision Tree, Random Forest and XGBoost for binary machine-failure prediction. The experimental results indicate that the XGBoost model performed best overall, with a precision of 0.661, recall of 0.725, F1-score of 0.692, PR-AUC of 0.660, and ROC-AUC of 0.965, thus detecting 37 out of the 51 failures in the held-out test set. In the experiment, Random Forest had the lowest Brier Score (0.022), which means it had better probability calibration. The results show the benefits of combining the machine-learning predictions, Digital Twin representation, and BI interfaces to convert operational IoT data into predictive risk data for maintenance decision making. It provides an AI-IoT-Digital Twin architecture that can be replicated and a set of empirical methods for predictive Business Intelligence.