Regional Risk Assessment of Bridge Inventories In California Using Machine Learning Techniques

Published

October 1, 2022

Author(s)

Jazalyn Dukes, Sujith Mangalathu

Abstract

Recently significant attention has been paid to the regional risk assessment of infrastructure systems as proper risk assessment can lead to informed decision and recovery strategies. This paper presents the regional risk assessment of bridge systems using machine learning techniques. Skewed two span bridges with skew angle ranging from 0 to 60 degrees is considered in this study. The selected bridge classes occupy more than 60 % of the California bridge inventory. Extensive numerical analysis has been carried out using the expected ground motions in California, and various machine learning techniques such as Lasso regression, Support Vector regression, K-nearest neighbors, Gradient boosting methods such as XGBoost, AdaBoost, and CatBoost has been evaluated in this paper. It has been noted that XGBoost outperforms other methods, and the selected model has a coefficient of correlation of 0.92 for predicting the damage in the unknown test set. The paper also identifies the significant factors that influence the damage assessment, and an easy-to- implement solution is suggested in this paper.

Conference Dates

July 5-7, 2022

Conference Location

Athens, GR

Conference Title

3rd International Conference on Natural Hazards & Infrastructure

Pub Type

Conferences

Keywords

regional risk, bridges inventories, machine learning, skew bridges, earthquake

Citation

Dukes, J. and Mangalathu, S. (2022), Regional Risk Assessment of Bridge Inventories In California Using Machine Learning Techniques, 3rd International Conference on Natural Hazards & Infrastructure, Athens, GR (Accessed December 3, 2024)