Comparing Domain Expert and Machine Learning Data Enrichment of Building Registry

Municipal decision-makers must define and quantitatively analyze full-renovation scenarios adapted to specific districts and buildings to achieve the European Union (EU) target of saving 60% to 90% of energy by renovating 75% of building stock. However, poor open-data quality presents a tenacious challenge, especially for automatic calculations or decision-making. This study addresses the challenge of enriching Estonian Building Registry (EBR) data by predicting the actual external wall type from existing registry information. To achieve this, both domain expert rules and machine learning models were employed. The study used a training dataset of 416 buildings and a test dataset of 66 buildings. While previous research comparing expert-based and machine learning approaches has been limited and yielded mixed results, our findings demonstrate that both methods perform similarly, improving the initial wall type classification accuracy from 54% to 89%.