A Bayesian probability model based on sub-market effects was proposed to address the situation that the hedonic price model HPM is prone to insufficient prediction accuracy and interpretability in the face of the complex relationship between house prices and characteristics. In improving the algorithm design the idea of sub-market clustering was borrowed a latent variable was introduced to represent sub-markets and sub-market criteria were established based on location proximity and substitutability. Next sub-market criteria and hedonic price models were used as probabilistic dependencies of Bayesian networks to determine the range of effects in each sub-market completing the sub-market segmentation. Finally house prices were predicted based on the probabilities of the submarkets to which the houses belonged and the key influencing factors of the submarkets were analyzed to improve the prediction accuracy and interpretability. The model was compared with five existing models in terms of mean absolute percentage error mean absolute error and root mean square error. The performance of the algorithms of the non-submarket model and the submarket model were tested separately based on property data of Hangzhou City before 2019. The experiments show that the Bayesian model outperforms the comparison models in terms of accuracy in forecasting real estate prices and has the advantage of interpretability.