Stock Market Trend Prediction Research Based On Feature’s Energy Calculation And Energy Deviation Analysis

The stock market is an important part of financial market and has a very important impact on national economic and social development. So, an effectively stock market prediction method has great significance for the masses of people. But because of Chinese stock market legal system is not sound, imperfect market mechanism and investor psychology factors, stock price prone to large fluctuations and is hard to predict.In this paper, aiming at solving stock price volatility, we analysis of stock price volatility from the perspective of energy, using support vector machine (SVM) model as forecasting model. Based on the historical data, we predict future price movements of the stock market. The main research work is as follows:K line feature is the causal information for the rise and fall of stock price, the current stock market forecasting model lacks of systematically analysis of K line feature. Based on K line energy calculation, a lifetime support vector machine (LPF-SVM) algorithm is put forward in this paper. First, the typical K line is extracted, with the introduction of maturity and explosive force definition, the K line feature of support vector machine algorithm (KLF-SVM) can be obtained. Then on the basis of KLF-SVM, the algorithm defines a typical energy calculation model, giving a kind of SVM prediction algorithm for K line energy calculation. According to the K line feature energy and the K line feature combination way to judge the current stage of stock price life, the algorithm predicts the fluctuation range of future stock price, taking the fluctuation range as prior knowledge to join the SVM. The experiment result shows that LPF-SVM algorithm has lower error than SVM algorithm and FWSVM algorithm.Because of the LPF-SVM is more suitable for the stock price which is relatively regular, in the case of irregular, the result is not ideal. We put forward a method (EDD-SVM) based on the deviation degree of K line feature energy and technical indexes energy. The algorithm build a bayesian network by close price and various technical indicators, and then extract the markov blanket of close price, searching the technical index which is most closely with close price. In order to predict the stock price effectively, we define the energy model of extracted technical index. We judge the range of stock price fluctuation according to the divergence of the technical index and the K line feature. Adding the range as priori knowledge to the SVM, the model is well to predict the stock price. The experiment result shows that EDD-SVM algorithm has lower error than LPF-SVM algorithm and KLF-SVM algorithm.