CONF Tognoli_BS2023_2023/IDIAP A Machine Learning Model for the Prediction of Building Hourly Heating Demand from CityGML Files: Training Workflow and Deployment as an API Tognoli, Marco Peronato, Giuseppe Kämpf, Jérôme 3D City Models CityGML Decision Tree Learning REST API Proceedings of Building Simulation 2023: 18th Conference of IBPSA 2023 2932 - 2939 https://publications.ibpsa.org/conference/paper/?id=bs2023_1570 URL 10.26868/25222708.2023.1570 doi We present a workflow for the development and deployment of a data-driven model to estimate the hourly heating demand of buildings.The model is trained and tested using CityGML with the Energy ADE specification and Meteonorm CLI weather-files as the source of the input features.Through an optimization pipeline, an ensemble model including a gradient boosting algorithm presenting a RMSE of 9.5 Wh/m2 of floor area (98.7% accuracy) and low memory requirements is selected. A short-term predicting model is also developed reporting a RMSE of 4.2 Wh/m2 of floor area (99.7% accuracy).A web service providing access to a REST API deploying the data-driven model is also developed, allowing for a wide range of applications in third-party tools such as in GIS analysis.