Accurate prediction of reservoir recovery rate is of paramount importance in the early stages of reservoir development planning. Conventional methods for predicting gas reservoir recovery rates are typically based on empirical comparisons or reservoir numerical simulations, both of which often exhibit poor accuracy in the early stages of reservoir development. This paper aims to conduct prediction research on the recovery rate of water-drive gas reservoirs using machine learning techniques. Firstly, relevant data from 70 water-drive gas reservoirs, including geological properties, reservoir recovery rate, etc., were collected, and a corresponding dataset was constructed. Based on feature extraction using principal component analysis to identify factors influencing recovery rate, machine learning algorithms such as Support Vector Machine, Random Forest, and LightGBM were employed to construct recovery rate prediction models. The models were evaluated, optimized, and their predictive performance and generalization ability were validated. A comparison between machine learning methods and traditional mathematical modeling approaches in predicting recovery rates of water-drive gas reservoirs was conducted, and the application prospects and limitations of machine learning in this field were discussed. The results indicate that machine learning-based methods can accurately evaluate the recovery rate of such reservoirs. This study provides a new prediction method for the development and management of water-driven gas reservoirs, demonstrating significant practical value and applicability.