Use clustering to improve neural network in financial time series prediction

In this paper, a time series prediction method using clustering to improve neural network is studied. The big data group is divided into some small parts by clustering. By this way, every small part has a higher conformity, and data in these small parts is used to train corresponding neural network for prediction. The prediction model is constructed from neural network with the addition of clustering and is applied to the financial time series prediction. The experiment results demonstrate the effectiveness of the improvement. Comparison with the primitive neural network prediction model shows that clustering increases neural network´s trend accuracy in continuous prediction, while debasing the cost of time and reducing the complexity of the prediction model.