The absolute open flow potential (AOFP) is defined as the gas flow rate when the bottom hole pressure is equal to atmospheric pressure, and this directly influences the productivity and appropriate allocation of production of shale gas well. The conventional method of determining the AOFP of shale gas wells is through a systematic well testing which is time consuming and has adverse economic effects. Therefore, developing an efficient and rapid intelligent decision method for predicting the AOFP is considered to be a win-win strategy for gas recovery as well as economic performance. In this study, one of the most classic unsupervised machine learning methods namely, principal component analysis (PCA), was combined with radial basis function neural network (RBFN), which is a low computational complexity deep-learning method (RBFN-PCA) in forecasting the AOFP for shale gas wells. This method includes PCA analysis for reducing the dimension of relevant geological and engineering factors and the RBFN model for nonlinear regression and prediction of the AOFP based on collected field data. The proposed method was applied to the Weiyuan reservoir in the southwest district of the Sichuan basin, southwest China. Eighty-four wells in the Weiyuan block were chosen as data samples for training and establishing the model with a relatively integrated set of data, including geological and engineering parameters as well as accurately recorded AOFP. Results showed that the proposed method could predict AOFP with high accuracy and efficiency if adequate data are accessible. Meanwhile, geological parameters including total organic carbon, brittleness index, gas content and porosity as well as engineering factors including proppant amount, flowback ratio, fractured length, and fracturing fluid injection rate were screened as the dominating factors that determine more than 85% weight of AOFP in the Weiyuan reservoir. The four model quality evaluation metrics showed that the performance of the RBFN model with the dominating factors only was better than using all geological and engineering factors. This indicated the superiority of combining the unsupervised-learning and deep-learning methods. More importantly, this proposed machine learning method provides an intelligent decision strategy for oil companies as well as researchers as a reference to solve similar problems in petroleum engineering.