IMPLEMENTATION OF MACHINE LEARNING TECHNIQUES FOR PROGNOSTICS FOR RAILWAY WHEEL FLANGE WEAR

Abstract

Machine learning has become an immensely important technique for automatically extracting information from large data sets. By doing so, it has become a valuable tool in various industries. In this investigation, the use of machine learning techniques for the production of railway wheel prognostics was investigated. Metrorail’s railway wheel wear data was used as a case study for this investigation. The goal was to demonstrate how machine learning can used on the data generated by Metrorail’s routine operations. Three machine learning models were implemented: logistic regression, artificial neural networks, and random forest. The investigation showed that all three models provided prognoses with an accuracy of over 90 per cent, and had an area under curve (AUC) measurement exceeding 0.8. Random forest was the best performing model, with an AUC measurement of 0.897 and an accuracy of 93.5 per cent.