Developing a Framework for Building Condition Assessment of Schools in Osijek-Baranja County
This study introduces a novel approach to building condition assessment (BCA) by combining traditional manual grading with machine learning models—artificial neural networks (ANNs) and random forests (RFs). Individual building components (e.g., windows, roofs, and floors) were assessed based on thei...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-04-01
|
| Series: | Buildings |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-5309/15/9/1511 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study introduces a novel approach to building condition assessment (BCA) by combining traditional manual grading with machine learning models—artificial neural networks (ANNs) and random forests (RFs). Individual building components (e.g., windows, roofs, and floors) were assessed based on their remaining useful life using an Excel-based system. The resulting total building grades were used to train and validate ANN and RF models. Performance was evaluated using <i>R</i><sup>2</sup>, mean squared error (<i>MSE</i>), root mean squared error (<i>RMSE</i>), coefficient of variation of <i>RMSE</i> (<i>CVRMSE</i>), and mean absolute percentage error (<i>MAPE</i>). The ANN model outperformed RF in the training set (<i>R</i><sup>2</sup> = 0.987, <i>MAPE</i> = 0.50%) and showed high accuracy in validation (<i>R</i><sup>2</sup> = 0.940, <i>MAPE</i> = 2.55%). The RF model also performed well (<i>R</i><sup>2</sup> = 0.942, <i>MAPE</i> = 2.66%), confirming its viability. External validation on data from outside Osijek-Baranja County confirmed model robustness, with ANN again achieving better performance (<i>R</i><sup>2</sup> = 0.799, <i>MAPE</i> = 7.71%) than RF (<i>R</i><sup>2</sup> = 0.747, <i>MAPE</i> = 9.17%). |
|---|---|
| ISSN: | 2075-5309 |