Developing a novel hybrid model for seismic loss prediction of regional-scale buildings

Seismic loss prediction of regional-scale buildings provides key information for disaster management. However, previous seismic loss prediction models ignored the time-lag effect of cost information and adopted rough loss ratios without considering the characteristics of a region and disaster (including GDP, magnitude, focal depth, epicentre intensity, seismic precautionary intensity, and design basic acceleration of ground motion), which prevented the earthquake loss prediction from achieving sufficient forecasting accuracy. This study develops a novel storey-based seismic direct loss prediction model of regional-scale buildings for the pre-earthquake hazard mitigation and post-earthquake disaster responses based on the machine learning approach. Specifically, the multi-factor support vector machine (SVM) model is established to predict loss ratios, and the hybrid model (ARIMA-SVM) composed of the autoregressive integrated moving average (ARIMA) model and SVM is built to forecast unit replacement costs. The results show that the above models can achieve promising forecasting performance. Finally, the prediction framework has been applied to Wangjiang Campus of Sichuan University in China, and the spatial distribution of storey-level seismic damage and losses in this region can be visualized. Importantly, the outcome of this study provides a comprehensive and realistic understanding of future losses in a region, which supports decision makings in disaster mitigation.