This paper explores a method for evaluating blended teaching effectiveness using machine learning. It constructs a blended teaching evaluation model with the random forest algorithm based on diverse data types and sources, including grades, behaviors, and emotions. Feature importance analysis is used to propose targeted strategies for teaching improvement. Using survey data from 376 students enrolled in the national top-level online course exploring geoscience landscapes: aesthetics and culture on Chinese University MOOC, the model is trained and validated. Results show that the optimized random forest model improves fit by 40%, reduces mean squared error by 7%, and achieves an R² of 0.92 when predicting student performance. Key factors influencing learning outcomes include participation, online-offline learning paths, mobile phone use, formative assessment, time investment, and sense of achievement. Based on these factors, targeted strategies for enhancing blended teaching effectiveness are proposed. The method provides references for optimizing blended teaching design, improving blended teaching quality, and promoting personalized education.