Effective prediction of medical training outcomes is of great significance for training institutions and organizations to optimize training strategies and resource allocation. The purpose of this study is to explore methods based on learner profiles and ensemble learning to improve the accuracy of predicting medical training outcomes. Traditional machine learning models may not perform well when facing nonlinear problems such as predicting the effectiveness of medical training. Therefore, this study applies the idea of ensemble learning and fuses multiple machine learning models through voting algorithms to better capture the complexity of training effects. To validate the model, real medical training data was used for experiments. Using 14 indicator data as input to the model, a series of learner profile labels were generated using statistical analysis and machine learning algorithms. These feature labels served as the input to classifiers, and prediction results served as output for the comparative model experiments. The experimental results show that the accuracy of the proposed model reached 83.94%, an average increase of approximately 2.4% over other traditional machine learning models and an average increase of approximately 7.4% over other ensemble learning models in the experiment, and its overall performance is better than that of other models, indicating that the application of this model can provide a more reliable basis for predicting the effectiveness of medical training.