Water vapor transmission rate (WVTR) is one of the most important indicators to evaluate the moisture retention of biodegradable mulch films. However, the multiple factors can dominate the dynamic evolution of WVTR during the effective service life of the film in practical production, including material composition, environmental parameters, and material properties. However, only a single variable cannot effectively quantify the multiple factors' interactions in the conventional experiment. It is often required for the specific contribution of individual factors to WVTR variations during aging. In this study, a systematic evaluation was conducted to predict WVTR using machine learning models. A database was constructed with 568 samples from published literature. 17 influencing factors were extracted for WVTR during aging. Four machine learning models—random forest (RF), support vector machine (SVM), categorical boosting (CatBoost), and extreme gradient boosting (XGBoost)- were developed to evaluate the WVTR prediction, according to the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Among them, XGBoost demonstrated the highest accuracy, with an R² of 0.88, RMSE of 103.86 g/m²·d, and MAE of 50.08 g/(m²·d). Shapley additive explanations (SHAP) was applied to quantify the importance of the 17 factors on the dynamic evolution of WVTR. The rainfall and ultraviolet (UV) radiation were the most significant factors with the largest contributions of each feature. Specifically, the increasing rainfall promoted water absorption, thus causing a more porous structure for the increasing WVTR. Conversely, UV radiation accelerated the degradation of polymer chains, leading to a higher WVTR, due to microcracks and polymer chain oxidation. Additionally, the properties of the films, such as thickness and area, directly dominated the WVTR. Thinner or larger film areas reduced vapor transmission resistance, thereby increasing the WVTR. The rest of the factors, such as the ratio of biodegradable components and the presence of additives, also influenced the water vapor barrier. Furthermore, 104 groups of experiments were employed to validate the adaptability and reliability in practical scenarios. Overall, the driving factors were determined to evaluate the prediction effectiveness of machine learning models, particularly XGBoost. Moisture retention and crop yield can be enhanced after more extensive field trials under diverse environments. The findings can offer a data-driven approach to optimizing biodegradable films in agricultural applications.