Deep learning classification approaches and applications for energy performance certificates (EPCs)

Building energy performance classification is a cornerstone of sustainable development initiatives. This study presents an innovative approach leveraging Artificial Neural Networks to classify Energy Performance Certificates. Our Artificial Neural Networks model, integrating the Synthetic Minority Oversampling Technique for class balancing and Principal Component Analysis for dimensionality reduction, achieved a test accuracy of 93.44%, supported by a macro and weighted F1-score of 0.93, outperforming many existing models and creating a unique sequence and combination of methods to conclude in that result. A detailed analysis of class-level performance underscores its robustness for high-rated energy classes while revealing challenges in differentiating lower-rated classes. This work bridges the gap between high-performance AI models and their interpretability, setting a benchmark for future energy performance certificate classification studies. • ANN model achieved 93.44% accuracy for Energy Performance Certificate (EPC) classification • SMOTE was applied for class balancing to address data imbalance issues • PCA was used to reduce dimensionality and enhance model generalisation • ANN outperformed existing models in classifying high-rated energy classes (A1–A4) • Combined use of SMOTE and PCA formed a novel and effective EPC classification pipeline