Archard model guided feature engineering improved support vector regression for rail wear analysis

•A physics model is applied to generate new features for the SVR model improvement when the raw data contains outliers.•A long-term rail wear degradation modelling and prediction approach is realized based on a 14-year in-field dataset.•A two-peak characteristic of rail wear was observed across the cant height value range.•Rail service time length is the most important factor for rail wear prediction.