Development of Random Forests Regression Model to Predict Track Degradation Index: Melbourne Case Study

In rail infrastructure maintenance management systems, Track Degradation Index (TDI) is considered as a representative of quality of rail tracks. This index is usually developed based on the deviation rate or standard deviation of track geometry parameters. In this regard, prediction of future TDI is an important task as it can be employed to determine when and where maintenance and renewal activities must be deployed. In this study, a track geometry data set from Melbourne tram network has been used as the case study and gauge deviation parameter is selected as the main parameter to develop TDI. For prediction of the future TDI, Random Forests (RF) model as a Machine Learning (ML) model is used to predict the future TDI of the data set. Since TDI is a continuous variable, Random Forests Regression (RFR) model is applied. In this study, RF model has added two algorithms to the basic Decision Trees (DT) model including bagging and random subspace method. These algorithms can reduce the overfitting problem and over-focus on special features. Based on the results of this study, adjusted R2 value of the proposed prediction model is 0.96, which demonstrates that the model has the satisfying performance in predicting the TDI.