The accurate prediction of production in tight gas reservoirs is a fundamental requirement for gaining a proper understanding of the reservoirs and developing rational development strategies. The prediction of production in tight gas reservoirs has always been a significant challenge due to the reservoir’s poor petrophysical properties, strong heterogeneity, and the complexity of fluid flow patterns induced by hydraulic fracturing operations. In this study, a multi variable time series production prediction model based on XGBoost regression is developed by constructing a sliding window to create a time series dataset. The training dataset, comprising the first 80% of the data, is utilized to train the model, while the remaining 20% of the data serves as the testing dataset to evaluate the performance of the model. By validating with data from five actual tight gas wells, the predictive performance of Arps decline analysis, the analytical model for limited drainage fractured well testing, and the multi variable time series production prediction model based on XGBoost regression (MT-XGB) was assessed using the root mean square error (RMSE) as the evaluation metric. The results indicate that the MT-XGB exhibits the highest predictive accuracy. Specifically, for gas wells with significant production fluctuations, this method significantly outperforms traditional approaches in terms of predictive accuracy. Moreover, for gas wells with relatively stable production changes, the MT-XGB still demonstrates superior predictive accuracy compared to conventional methods. Overall, the MT-XGB accurately captures the dynamic variations in daily gas production.