Stock forecasting is an important part of stock market investment management. In this study, stock was predicted using data mining method, and back propagation neural network (BPNN) was taken as the basis. Genetic algorithm was used for optimization to obtain the improved BPNN algorithm, and then it was applied to stock prediction. Instance analysis was performed taking a stock as an example. The forecasting results of BPNN, particle swarm optimization (PSO) improved BPNN and the improved method were compared. The results showed that the forecasting results of the improved BPNN method were basically consistent with the actual values, with an average error of 0.8%, much lower than 7.19% of the BPNN algorithm and 4.17% of the PSO-BPNN algorithm, suggesting that the improved BPNN algorithm was valid. Then four stocks were predicted to understand the future development of those stocks and help select proper stock for investment. It suggested that the improved BPNN algorithm could provide a reliable basis for stock selection. This study provides some theoretical support for the further application of data mining in stock forecasting, which is helpful for investors to make correct stock choices, improve returns and avoid risks. Moreover, it also contributes to stabilizing the stock market and promoting economic development.