Improvement of Rail Track Degradation Prediction Models by Detecting Outliers and Enhancing the Dataset

In rail infrastructure, modelling the degradation prediction of rail track is an important step toward developing preventive maintenance strategies. Using quality data anddatasets without the least error and noises can improve the accuracy and reliability of the models’ predictions. In this study by the application of two outlier detection methods including Interquartile Range (IQR) and Median Absolute Deviation (MAD), datasets required for the development of track degradation prediction models have been prepared. In this research, data of tram track gauge from Melbourne’s tram system have been used. Based on each outlier detection technique, different datasets have been created. For prediction modelling of the tram track degradation, Artificial Neural Network (ANN) algorithm has been used. The results of the study show that the models based on IQR and MAD filtered data can provide more reliable forecasts than the model based on the unfiltered data.

Keywords:

Median Absolute Deviation , Interquartile Range , Tram Track Degradation , Melbourne tram