A Framework for Rail Surface Defect Prediction Using Machine Learning Algorithms

Defects on the rail surface may cause damage to both vehicle and the passengers who are travelling in the vehicle or to the goods in freight trains. So, predicting these defects at the earlier stage may prevent the rail surface from catastrophic failures. Defects in the rail surface are either due to some external factors or it may be because of the impact of the passing train. In this paper, an analytical study of rail surface defect prediction using three machine learning algorithms namely Neural Network, Decision Trees and Random Forest has been carried out in R. Here, the data is also pre-processed because under some speed levels the defects are not recorded. These three algorithms can learn from data provided and can make predictions with testing data by building a model from training data. The algorithm, which have the highest accuracy for a particular defect can be chosen for that defect prediction for other data. Defects are predicted to be at low defect or high defect.