In order to study how to accurately predict the development trend of track irregularity in the absence of historical data， a track irregularity prediction method that can consider the prior information of experts is proposed. The questionnaire survey method was used to obtain expert experience information and a Bayesian linear regression model with prior information was built up. Then the Markov chain Monte Carlo method was used to solve the model parameters. Finally, the amplitude of track irregularity was predicted and error analysis was conducted, and the prediction effects of different models in the absence of historical data were compared. The results show that the method can accurately predict the development trend of the track irregularity in the short term, and the correlation coefficients are all above 0.9. In the absence of historical data, the Bayesian linear regression model can also maintain a high prediction accuracy and the R2 is 0.88, which is 17% higher than the traditional linear regression model.