Prediction of Share Price Trend Using FCM Neural Network Classifier

High-noise, chaos, non-linearity and instability are notable features of share price time series. Traditional economic model assumes that the change of share price is linear, but the assumption does not conform to reality. Therefore, the accuracy of prediction of traditional economic model is not satisfying. In this paper, considering these existed problems of traditional model, a novel method, Floating Centroids Method (FCM), is used to establish the share price trend model. FCM algorithm fits law of share price trend by finding the optimal neural network. Through the optimal neural network, share price data points are mapped into a new space which is called partition space. In the new space, same tendency of share price data points are as close as possible and different tendency points are as far as possible. Then, share price data points are clustered by K-means algorithm in partition space. Every cluster is classed. Lastly, the class of the cluster that share price point belongs to is taken as share price trend in the future. Based on experimental data, FCM algorithm has higher average accuracy and better generalization ability than comparative algorithms.