A Research on Neural Network and An Improved Model For Stock Prediction

Abstract

In this paper, Back Propagation Neural Network (BPNN) and Firefly Algorithm-Back Propagation Neural Network (FA-BPNN) are studied for the purpose of stock price prediction. The result shows that, compared with BPNN, the average prediction time of FA-BPNN increases by 3.2239s, the average relative prediction accuracy increases by 1.2739 %, and the corresponding variance decreases by 0.0003. It shows that the prediction accuracy obtained using FA-BPNN is higher. This is because the weighted value of FA-BPNN can effectively find the extreme value of the loss function, thus avoiding local optimum. But the prediction time is slightly longer, because the algorithm needs to sacrifice the tracking time in order to improve the prediction accuracy. Meanwhile, the variance corresponding to the relative prediction accuracy of FA-BPNN is the smallest. This demonstrates that the prediction effect of FA-BPNN is superior, and the prediction results are more stable and robust after each training. The above outcomes demonstrate that the FA-BPNN model has excellent performance in short-term stock forecasting. It has some feasibility and can be of use to stock market investors.