Reading enjoyment is crucial to students' reading performance and reading behaviors. A key concern for educators is identifying the factors that predict children's reading enjoyment. However, previous research has predominantly employed traditional regression methods, which may oversimplify the complex associations between predictors and reading enjoyment. In this large-scale study, we investigated the impact of demographic variables, reading motivation, and the family reading environment on reading enjoyment among 10,979 Chinese children participating in a summer reading program. We utilized five machine learning algorithms (Linear Regression, Random Forest, Gradient Boosting, XGBoost, and Support Vector Regression) to assess the contribution of 18 predictors to reading enjoyment. The XGBoost model was selected as the optimal model, demonstrating strong predictive performance (R² = 0.6795, MSE = 0.3944, RMSE = 0.6280, MAE = 0.4434). Subsequent SHapley Additive exPlanations analysis (SHAP) revealed that the five most important predictors of reading enjoyment were intrinsic motivation, reading persistence, social motivation, mastery goals, and self-efficacy, all of which positively predicted children's reading enjoyment. In contrast, performance-avoidance goals and performance-approach goals were found to be negative predictors. Further stratified SHAP analysis indicated that the SHAP value for parental education expectation was the only feature to show a significant difference between boys and girls (difference = 0.001, p = 0.003). Beyond grade level and age, the SHAP values for nine features differed significantly between lower-grade (Grades 1-3) and upper-grade (Grades 4-6) students. Specifically, personal book collection, intrinsic motivation, self-efficacy, and performance-approach goals were more influential predictors for upper-grade students (ps < 0.05). Conversely, parental reading support, social motivation, mother's educational level, parent-child reading activities, and personal education expectation had a greater predictive contribution for lower-grade students (ps < 0.05). These findings provide robust empirical evidence to inform targeted interventions designed to enhance children's reading enjoyment.