Structural damage evaluation in RC buildings through ensemble learning: A comprehensive study of different techniques for efficient and reliable identification

Abstract

In the field of damage detection, the integration of machine learning (ML) techniques, particularly Ensemble Learning (EL), has proven to be a robust method for effectively processing large datasets from various sources. As the use of EL expands in this area, it becomes crucial to evaluate the relative effectiveness of different EL approaches. Comparing these methods provides key insights into the data and establishes benchmarks for evaluating techniques. This study investigated three EL classifiers: Random Forests (RF), Gradient Boosting (GB), and Bagging. The focus was on first, discovering the potential of ML methods, especially EL algorithms in classifying damage based on experts’ survey in reinforced concrete buildings in Nepal, Ecuador, Haiti, and South Korea then predicting damage levels. Additionally, a novel index called the Probabilistic Uncertainty Measure (PUM) was introduced to improve the interpretability and reliability of the EL-based results. This index assesses the probability of misclassifying damage categories, offering a refined perspective on the dependability of EL findings. The research highlights that the Bagging and RF classifiers significantly outperform others, with accuracy improvements of 73 % and 67 %, respectively. Furthermore, the PUM index shows that Bagging and RF consistently deliver reliability values 84 % and 83 % higher than other methods, confirming the strong potential of EL techniques in accurately identifying damage in reinforced concrete buildings.