Model of Predicting the Rating of Bridge Conditions in Indonesia with Regression and K-Fold Cross Validation

Abstract

Maintenance and repair of the bridge are inevitable in the operation of a bridge to maintain its condition to keep the operation. Indonesia has hundreds of thousands of bridges that are still actively in use. The classic problem with infrastructure management, such as bridges, is that large numbers are generally not balanced with adequate bridge maintenance budgets. Therefore, the strategy of implementing maintenance and repair by preparing priorities becomes the only logical approach. To get a priority scale, a scoring mechanism is needed. The assessment used by the Ministry of Public Works and Public Housing (PUPR) especially the Bina Marga field is based on the bridge management and maintenance system, namely Bridge Management System (BMS) 1993. With BMS 1993, the condition of the bridge is represented by the Condition Value (NK) of the bridge. This study is based on existing NK, prediction of NK value in the future. The predicted model developed is with regression models. Regression models are combined with k-fold cross-validation to improve the accuracy rate of the model. The developed model produces regression models for all variables of condition values with a low error percentage that is in the range of MAPE = 10% and RMSE 0.15. Further significance tests with ANOVA are also conducted to test the effect of independent variables on dependent variables, including testing on fit models to show the resulting model does not overfit and/or underfitting.