In order to predict the degree of water stress of tomato in greenhouse, sensors were used to obtain the internal environmental information of greenhouse, including air temperature (Ta), air relative humidity (Rh), substrate humidity (Hs), light intensity (Li), carbon dioxide concentration (CO2) and substrate temperature (Ts). The wind speed (Ws), outdoor relative humidity (Rho) and outdoor air temperature (Tao) of the greenhouse were obtained from local weather station. According to the above nine parameters, the crop water stress index (CWSI) prediction model of greenhouse tomato was established based on CS-CatBoost. The feature weights were calculated and screened by the gradient lifting algorithm. The performance of the CS-CatBoost algorithm under different input feature numbers was compared with the original CatBoost model, CS-LightGBM model and CS-RF model. The results showed that when the number of input parameters of the model was 7, compared with CatBoost, CS-LightGBM and CS-RF, the RMSE was decreased by 0.0123, 0.0118 and 0.0311, MAE was decreased by 0.0066, 0.0075 and 0.0208, MAPE was decreased by 0.9630, 1.1232 and 3.0892, while R 2 was increased by 0.0177, 0.0165 and 0.0767. When the number of other parameters as the model input, CS-CatBoost models prediction ability was superior to the other three model. The research result proved that the CS-CatBoost model had better prediction ability and generalization ability, which provided a strategy for water stress degree analysis of greenhouse tomato cultivation, thereby improving the utilization efficiency of agricultural water resources.