Research On The Prediction Of Stock Prices Trend Based On K Nearest Neighbor

Stock can be widely gather the idle funds of social, which is not only can increase the scale of the national economic construction, but also can promote the capital transverse accommodation, the economy of horizontal linkages and improve the total benefit of the resources allocation. The change of stock price not only can reflect the development of domestic economy and the running status of company, but also the basis of enterprise financing and stock investment. Therefore, predicted the price of stock rise and fall can overcome the blind investment effectively for the majority of investors.This paper combined with the existing problems of the stock trend both at home and abroad related research in the study of dynamic, in order to achieve the purpose of dimension reduction, the statistics are used to fusion the characteristics, which is describe the stock, So the new features of the stock space are constructed. According to analysis the past characteristics data of stock,judgement the existing stock price trends. The main work of this paper is as follows:Firstly, Analysis and determine the characteristics and trends of stock. Based on analysis of all kinds of stock price index, this paper selects a certain period of the opening price, closing price, the highest and the lowest price, the total hand, turnover rate, amount and volamount as the characteristics which can described the stock, and then take rising trend and the falling trend of stock price as category labels, based on this, the prediction problem of stock is converted into two classification problem.Secondly, the fusion and dimension reduction the characteristics of stock. Extracting 15 days data of each characteristic and 150 features are got. These data are easily lead to large computation and judgement result is not accuracy. In order to achieve the purpose of dimension reduction, the statistics are used to fusion the characteristics, such as the mean, standard deviation, median, kurtosis and skewness, so the new features are got and the dimension reduced from 150 to 50. The training set and testing set can obtained by this method.Again, The K nearest neighbor classification method is applied to the predicted the trend of stock price and given the detailed steps.At the last, the training set and testing set are constructed by fusion and dimension the five industries history transaction data from A-share market(such as computer application industry, Chinese medicine industry, food processing and manufacturing industry, biological products industry and communications equipment industry). In every industry data, taking 200 groups of rising trend segments and 200 groups of falling trend segments, let the rising trend characteristics of stock data as the first category, and the falling trend characteristics of stock data as the second category, respectively, selected two thirds before of two classes of the total data as a training set, a third after as the test set, take K equal to 15, the results of five industry classification accuracy are 100%、99.25%、99.00%、99.75% and 98.75%, respectively. The results of Tong Huashun simulation platform is the short-term prediction of the trend in the four stocks are accurate except one stock is influenced by the rights issue.Experimental results show that the pattern recognition classification technology to the stock price trend analysis is feasible. But the good news also has a certain effect on the growth trend of the stock, therefore, The fundamentals of information will fusion in the future research, further do the combining qualitative analysis with quantitative analysis.