Applying random forest algorithm for highway bridge-type prediction in areas with a high seismic risk

Machine learning was used to bridge-type prediction in the preliminary design. Based on the decision tree, in the United States, the accuracy of highway bridge-type prediction showed to be until 95% in the areas with a low seismic risk, whereas it proved to be lower than 80% in the areas with a high seismic risk. This work improved the accuracy of highway bridge-type prediction in states with a high seismic risk, i.e. California, Nevada, Oregon, and Washington. Bridge basic datasets from these states were elaborated from the National Bridge Inventory database using the random forest algorithm, which was found to be better. Indeed, the average recall rate of accuracy of highway bridge-type prediction across these states was 85.4%, i.e. 5.8% higher than the most accurate recall rate according to the literature review. Additionally, this work determined the most important parameters of highway bridge-type prediction using sensitivity analyses. According to comparisons with some predictions based on important parameter combinations, the accuracy in California, Nevada, Oregon, and Washington was improved by a percentage of 10.9%. Moreover, the analysis of the 'year built' parameter underlined its importance. Determining bridge-types was improved, thus enabling the accumulation of technical experience and saving labor and time.