Geometry-Driven BIM Element Category Recognition from
Open IFC Property Records: A Reproducible Engineering
Science Study
Sonia Ahmed1,∗, Marek Salamak2
1Management in Construction Dept., CTU, Czech Republic
2Civil Engineering Dept., Silesian University of Technology, Gliwice, Poland
Emails: soniaahmad88@gmail.com; marek.salamak@polsl.pl
Abstract
Accurate object semantics are essential for building information modeling (BIM) workflows
to enable interoperability, model checking, quantity take-off, performance analysis, and other
downstream engineering applications. However, in practice, Industry Foundation Classes (IFC)-
based model exchanges often feature limited or poorly identified semantic tags, particularly dur-ing
interoperability with authoring and reviewing tools. This research proposes a re-producible,
geometry-based learning algorithm for the automatic recognition of BIM element categories
based on publicly available IFC-based property data. The empirical analysis is based on 780 ob-ject
instances from ten BIM categories from a publicly available sample of IFC object records. A rulebased
parser translates semi-structured BIM text exports into engineering features as boundingbox
dimensions, coordinates, elevations and object-status. The study compares three supervised
machine-learning baselines via stratified five-fold cross-validation: logistic regression, random forest
and extra trees. Random forest performed best overall with an accuracy of 0.992, balanced accuracy
of 0.971, a weighted F1-score of 0.992, and a macro F1-score of 0.970. The analysis of feature
importance shows that bounding-box height, width, length, spatial coordi-nates and externalityrelated
descriptors are the most important features. The results demon-strate significant
semantics can be extracted from minimal engineering descriptors without the need for deep
learning of meshes. This work provides an interpretable and efficient baseline for BIM enrichment,
assessment, and interoperability-focused preprocessing for engineering science use-cases.
Keywords: BIM; IFC; Semantic enrichment; Engineering informatics; Building data analytics;
Machine learning