Evaluating Energy Performance Certificate Data with Data Science

The related problems of improving existing buildings' energy performance, reducing energy consumption, and improving indoor comfort and their many consequences are well known. Considering increasing urbanization and climate change, governments define strategies to enhance and measure buildings' energy performance and energy efficiency. This work aims to contribute to the improvement of buildings' characteristics by conducting a thorough systematic literature review and adopting a data science approach to these problems, presenting initial results with an open-access energy performance certificate dataset from the Lombardy Region, in Italy. We provide a pre-processing method to the data, applicable for future research, aiming to address challenges such as automatic classification of existing buildings' energy performance certification, and predicting energy-efficient retrofit measures, using machine learning techniques. The analysis of this dataset is challenging because of the high variability and dimensionality of this dataset. For this purpose, a robust iterative process was developed. First, the data dimensionality was reduced with Pearson Correlation to find the best set of variables against the non-renewable global energy performance index (EPgl, nren). Then, the outliers were handled by utilizing Box Plot and Isolation Forest algorithms. The main contribution is to inform private and public building sectors on dealing with high dimensional data to achieve enhanced energy performance and predict energy-efficient retrofit measures to improve their energy performance.