Evaluating Four Types of Data Parsing Methods for Machine Learning Integration from Building Information Models

A method and structure for architectural datasets is proposed, specifically designed for analysis, sorting and ultimately reusing building elements. Four different methods of parsing data from real-life projects, using their building information models (BIM) for integration into a Machine Learning (ML) model, where evaluated. As ML integration is becoming more important in the AEC industry, we see an increasing demand on high quality datasets. Four different methods and file formats where benchmarked focusing on read and write speeds for converting architectural BIM models into datasets to be used in ML. Our results show that the current way of storing our projects in Industry Foundation Classes (IFC) is not optimal for development and integration of new AI assisted tools. This paper provides alternative methods and storage solutions for both developing new datasets internally but also for future work in creating a common federated learning setting for the AEC industry.