Novel ensemble learning approach for BIM object classification in the building performance domain

Duygu UtkucuHuaquan Ying, and Zijian Wang, at Technion – Israel Institute of Technology led by Rafael Sacks developed a novel integrated approach for the classification of BIM objects for building performance evaluation, which now has been published in the Journal of Advanced Engineering Informatics.

In this article, we focused on the automated classification of architectural and MEP BIM objects to address data exchange and interoperability challenges in pre-processing building models for performance evaluation. We suggested a series of ensemble models using a stacking mechanism of four classification tools. These four tools exploit different object features, as follows:
– Semantic Keyword Search (SKS) Tool utilizes object alphanumeric properties.
– Rule-based Inferencing (RBI) Tool utilizes spatial topology relationships between objects.
– Machine Learning (ML) Tool utilizes 3D object geometry features.
– Deep Learning (DL) Tool utilizes object visual shape features.

In addition, we presented a BIM object dataset comprising 3,410 instances across 40 classes and introduced eight ensemble models. The best ensemble model, applying all four classification tools, achieved 91.0% prediction accuracy. This demonstrated that automatic classification using ensemble learning is an effective strategy.

Here is the link to the full article:
Free to access by May 15, 2024:

Download the article dataset:

A Milestone Achievement: VCLab’s Research Paper Wins Prestigious Thorpe Medal

We proudly announce that a paper penned by our researchers at the Seskin Virtual Construction Laboratory (VCLab) has been awarded the esteemed Thorpe Medal by the European Council on Computing in Construction (EC3). The Thorpe Medal recognizes a paper contributing to practical or research aspects of engineering informatics disciplines in the built environment. This award underscores the practical value and impact of the paper on engineering informatics practice.

Prof. Rafael Sacks (right) and Zijian Wang (left) receive the Thorpe Medal from Dr Pieter Pauwels (middle) at the 2023 EC3 & CIB W78 Conference in Crete, Greece

Titled “Exploring graph neural networks for semantic enrichment: Room type classification,” the paper was authored by two PhD Candidates, Zijian Wang and Timson Yeung, and the Head of the National Building Research Institute and the VCLab, Prof. Rafael Sacks.

Team photo (From left to Right, Prof. Rafael Sacks, Zijian Wang, Timson Yeung)

This groundbreaking paper pioneers the use of graph neural networks (GNNs) for semantic enrichment of Building Information Modeling (BIM) by compiling BIM models into graphs. The work developed a novel graph dataset called RoomGraph for room type classification task in the scenario of residential apartments and improved a GNN algorithm, SAGE-E, to leverage edge features as well as node features. Compared to other GNN and machine learning algorithms, SAGE-E showed higher accuracy and balance in prediction while also shortening the training and validation process. Both the dataset and the algorithm are open for public research use at From the knowledge contribution aspect, this study validates that the BIM models can be represented as graphs and implements advanced GNNs to enrich BIM graphs. The success of the study will inspire researchers to explore the application of GNNs to BIM graphs for more complex intelligent functions.

Framework of the work

This recognition greatly highlights Technion’s continual contribution to the digitalization of construction and positions it at the forefront of AI applications for the AEC industry. Additionally, this study is part of the Cloud-based Building Information Modelling (CBIM) project, receiving funding from EU Horizon 2020 Marie-Curie grants. The award exemplifies the success of the CBIM EU training project in educating the forthcoming generation of BIM experts. Its focus includes automating the creation and enrichment of digital twins, enhancing the management, security, and resilience of BIM-enabled processes, and fostering wider BIM adoption across various sectors.