Financial Prediction Models Based On Data Mining

Financial prediction is an important research field in financial data mining. Besides being nonlinear, non-stationary, and dynamic, financial time series also has special properties, being high noisy, non-normal, sharp-peaked and heavy-tailed. So, financial prediction is more challenging, and has great values in practical application and bright prospect in marketing.This thesis investigates the applications of fuzzy revising model and a hybrid model based on ANN and feature extraction clustering in financial prediction. Because of the limitation of the original fuzzy logic model, when using it for trend forecasting, the trend accuracy ratio is low and the consecutive predicting values fluctuate flatly, which cannot reflect the real tendency of market. In order to overcome such deficiency, a fuzzy revising method is presented. The experimental results show that using fuzzy revising model for financial prediction is effective and feasible. In the hybrid model base on clustering and ANN, this thesis investigates the method of subsequence clustering in financial time series based on features extraction, and applies the results to train ANN, eliminating the bias of irrelevant history patterns. The experimental results show that the hybrid forecasting model outperforms traditional BP network in trend accuracy.Based on the same experimental data, the thesis compares the results of these two new forecasting models, and analyzes the respective characters and applicability.Finally, the thesis makes a research on how to choose the parameters of the forecasting models, and presents some simple but effective methods.