Building defects impact aesthetics, sustainability, functionality, conditions, performance, user satisfaction and safety in facilities. Despite an extensive catalogue of building defects, defects continue to increase unabated. Managing these defects is often tedious, costly, problematic and time-consuming, requiring significant resources from both the maintenance organizations, clients and stakeholders. This study aims to leverage machine learning to classify and predict defects in buildings.
A data set of 300 images, covering defects in four building components, was used to train the model for defect classification through machine learning. Four ensemble algorithms were trained for the predictive models.
The stacked model outperformed other learners in predicting defects across diverse data sets. Gradient Boosting achieved an accuracy of 76%, AUC of 92%, precision of 0.78, recall of 0.76 and F1-score of 0.75. The stacked model excelled with an accuracy of 83%, AUC of 94%, precision of 0.83, recall of 0.83, Matthews correlation coefficient of 0.75 and F1-score of 0.83, particularly in classifying defects in walls and columns.
These findings have theoretical and practical implications for real estate management research. This study contributes to building maintenance management knowledge by introducing a novel ensemble learning framework for robust defect prediction.
