Automated Machine Learning Techniques in Prognostics of Railway Track Defects

Abstract

The readiness and usefulness of Automated Machine Learning (AutoML) methods in classification of railway track defects is explored. Safety of railway networks is the top priority in the railroad industry, and track defects are a common cause of train accidents and service disruptions around the world. Effective classification and prediction of these defects based on historical inspection data can help in planning maintenance activities before critical defects occur. This increases safety of the network and lowers costs of the maintenance. The experimental analysis carried out on data from an international predictive modelling competition has shown that the proposed AutoML approaches resulted in an improved performance in comparison to the competition winning solutions and have an excellent potential for building robust predictive models in railway industry.