Railway track vertical irregularities classification based on vehicle response using machine learning method

Abstract

The common method of railway track irregularities assessment currently used, based on track geometry measurement data, is expensive to run, disrupts regular train services, and shows poor correlation with the actual vehicle response when running on a given track section. Vehicle response-based track irregularities assessment methods aim to overcome the shortcomings of geometry measurement-based methods. One of the possible methods is to establish the correlation between track geometry and vehicle response statistically using machine learning. Classification algorithms have been used in previous research for categorizing track sections into discrete classes. In this paper, in addition to discrete classification, the probability for which a track section is categorized as a certain class is presented. Testing results of the trained machine learning classification model showed high accuracy, precision, and recall. The results indicated that this method is suitable for classification of the vertical track irregularities. Further testing was performed with shifted classification threshold. The finding showed that the classification model’s precision and recall could be adjusted by shifting the classification threshold. The result of this study highlights the potential of machine learning classification usage for track irregularities assessment.