Determination of key variables influencing the daily gas production in the Sulige tight gas field using machine learning methods

China is one of the major countries blessed with abundant unconventional oil and gas with a recoverable tight gas reserve. To identify the critical factors affecting gas production in the Sulige tight gas field in China, a data-driven methodology is presented in this research. In this study, daily production data from vertical wells in the Sulige tight gas field, along with geological and completion data, were gathered and evaluated. By employing different machine learning algorithms, the importance of the individual variable was ranked, and a meta-rank was later developed by combing the results of the individual ML rankings. The most crucial factors influencing gas production in the area were identified using a variety of machine learning feature selection techniques. The results show that the injection pressure and Injection rate are at the back rank for the Random Forest model and the Extreme Gradient Boosting Model. The variables ranked as important were both the same which shows that our model is reliable although the relative importance may differ for the individual models. Furthermore, from the Boruta model, we could see that only seven variables were accepted as important after ninety-nine iterations. However, it produced a similar output for the least important variables. The Boruta Model rejected the Injection rate and Injection Pressure as important. The findings of this study are particularly important for examining the field's production performance considering shifting geological and completion parameters. The recommendation from this study can be applied to the design of new wells in the field and other similar tight gas reservoirs.