Validating prediction accuracy of an integrated deterioration method using the NBI dataset

Abstract

An integrated deterioration method incorporating time-based model, state-based model with Elman Neural Network (ENN) and Backward Prediction Model (BPM) currently has been developed. The aim of this method is to effectively predict long-term performance of bridge elements for various situations in terms of the quantity and distribution of available condition rating data. In order to validate the prediction accuracy of the proposed integrated method, this study is employed U.S. National Bridge Inventory (NBI) dataset as input to predict condition ratings of bridge components. The predicted outcomes are validated by a cross-validation method including backward and forward validations. The outcomes also demonstrate that the time-based model provides more accurate prediction outcomes than the state-based model when the given inspection records are sufficient.