Hybrid optimized RF model of seismic resilience of buildings in mountainous region based on hyperparameter tuning and SMOTE

Abstract

This study aims to develop hybrid-optimized random forest (RF) model of seismic physical resilience evaluation of buildings in mountainous region. Based on the earthquake-damaged building inventory by field survey in Shuanghe Town, the epicenter of Changning Ms 6.0 earthquake on June 17, 2019, 19 factors including seismic, geological, topography, environmental and building attributes were selected as the conditioning factors of seismic physical resilience evaluation of buildings to establish a database. The dataset was randomly divided into training and test dataset by 6:4. Based on training dataset, after hyperparameters tuning and unbalanced data processing, RF was used to develop evaluation models for buildings seismic performance. Finally, the confusion matrix, accuracy, precision and recall of the test dataset-based prediction evaluated the model's performance. The results shown that: The accuracy of model prediction were 0.88 and 0.83 for the training and test dataset, respectively. Hyperparameters tuning decreased the model's overfitting. Unbalanced data oversampling increased the accuracy, precision and recall obviously. The proposed hybrid optimized-RF model shows considerable stability and excellent performance in the seismic physical resilience evaluation of buildings in the study case. This could provide a novel reference method framework for improving the evaluation efficiency of earthquake-damaged buildings in mountainous regions in the future.