Hybrid machine learning model for prediction of vertical deflection of composite bridges

Abstract

In the present study, we have developed a novel hybrid Machine Learning (ML) based model namely B-IBk which is a combination of Bagging (B) ensemble and Instance-based K-nearest neighbors (IBk) predictor, for quick and accurate prediction of vertical deflection of steel-concrete composite bridges. In the models’ study, we have used five easily determined input parameters: cross-sectional shape, length of concrete beam (m), number of exploitation years, height of main girder (m), and distance between the main girders (m) to obtain output parameter: maximum vertical deflection (mm). For the development of models, direct measurement data of 80 steel-concrete composite bridges located at different places in Vietnam was collected and used as input and output parameters. Standard statistical evaluation indicators namely Mean Absolute Error (MAE), Correlation Coefficient (R), Root Mean Square Error (RMSE) were used to validate and compare the models’ performance. Results indicated that performance of the novel hybrid model B-IBk is very good (R = 0.908) for the prediction of Y of steel-concrete composite Bridge and better than single IBk model (R = 0.875) on testing dataset. Therefore, the developed novel model B-IBk is a promising tool for the accurate prediction of Y of Steel-Concrete Composite Bridges.