This study plans to collect 17 open-end balanced stock funds data from websites of domestic securities companies, selects the funds with the technical efficiency value of 1 as investment targets by using data envelopment analysis to analyze fund performance. Then, the mutual fund net worth prediction model is built by various new data mining methods including Backpropagation Neural Network (BPN), and the forecasting ability is compared with the Multiple Regression model. Through RMSE, we can understand the pros and cons of these fund forecasting models. This result is available for reference to investors as an investment strategy.