Damage prediction for regular reinforced concrete buildings using the decision tree algorithm

• A machine learning algorithm is used to develop damage predictor tool.
• Qualitative meaning of damage instead of quantitative representation is used.
• A few earthquake characteristics and structural properties are required as the input.
• Accuracy of the tool is verified using earthquake portfolios and numerical analysis.

To overcome the problem of outlier data in the regression analysis for numerical-based damage spectra, the C4.5 decision tree learning algorithm is used to predict damage in reinforced concrete buildings in future earthquake scenarios. Reinforced concrete buildings are modelled as single-degree-of-freedom systems and various time-history nonlinear analyses are performed to create a dataset of damage indices. Subsequently, two decision trees are trained using the qualitative interpretations of those indices. The first decision tree determines whether damage occurs in an RC building. Consequently, the second decision tree predicts the severity of damage as repairable, beyond repair, or collapse.