Fostering Understanding: Bridging the Gap Between Black-Box Models and Human Interpretability with Explainable Artificial Intelligence

The energy industry has benefited the most from the deployment of artificial intelligence (AI) techniques, particularly Machine Learning frameworks, for several purposes, such as assessing construction energy functionality. Nevertheless, the black-box models might end up in results that are difficult to human interpretability, making it impossible to use AI effectively in certain real-life circumstances. To improve the explanation of the system's findings, explainable AI (XAI) technologies may be successfully integrated into an AI-powered energy analysis approach. This study presents an XAI-based benchmarking system for predicting the membership to particular energy functioning categories of an extensive number of flattened energy-performance certifications (EPCs). Several black-box models with excellent accuracy are used to produce the categorization; however, their inference method is not accessible to humans. To further comprehend the model's conduct and the reasons behind both accurate and inaccurate categorizations, a universal XAI technique is used. This approach combines a regional explanation with a clustering technique for clarifying the outcomes of the model and the causal relationships between the predictions and the intended factor. The study offers a broad empirical strategy that can make utilization of a restricted number of examples to gather, clarify, and understand inference processes that the model has learned and finds valuable for the user. The system has been evaluated on around 100,000 EPCs of apartments in Italy.