Font Size: a A A

Research On Stock Trend Prediction Based On Behavioral Feature And Fractal Dimensionality Reduction

Posted on:2016-01-16Degree:MasterType:Thesis
Country:ChinaCandidate:Y H ChenFull Text:PDF
GTID:2349330503486890Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
With the rapid development of machine learning and artificial intelligence, which is increasingly used in the field of stocks on financial market. Most of researches use stock index in various kinds of prediction model. Stock index is a statistical category which utilize mathematical statistical methods. Stock index may demonstrate the tendency of stock price. There are considerable amount of stock index in Securities market. Some new behavior features of stock index are presented from extracting index in this dissertation, and these features will be used to predict stock price trend.There are many related researches put the stock index into machine learning model as features to predict the stock price trend, however, most of those researches just focus on the prediction model without more processing of these features, even no consideration of the usage on the stock index.In order to improve the learning efficiency and the precision of stock price prediction, this dissertation uses statistic analysis method to optimize the stock index as the features in training model, and using fractal dimension reduction algorithm to find the redundancy of those index features. The numerical part of behavior oriented indicators makes nonsense, such as KDJ indicators, only behavior part of the index will show the specific meaning. For such indicators, this dissertation adopts the quantification method to achieve the optimization of index., The experiments in this dissertation show that the optimized indexes have a better performance.In terms of index selection, the traditional dimension reduction methods such as singular value decomposition(SVD) may destroy the original data set after its selection processing, meanwhile it`s difficult to analyze the data after the dimension reduction.So on particular data set of stock index feature, this dissertation improves a dimension reduction algorithm based on fractal dimension which is adapted to the large-scale stock index data set. The dissertation proposed a new filtering method based on the reduction algorithm.Through experimental comparison, the new fractal dimension reduction algorithm shows better performance, also the reduction results is similar to SVD`s, even more the new method can keep relatively important features of stock.This dissertation collected more than 2000 stocks` daily trading data. After using the quantitative indicators and the new index selection method, the support vector machine perform better in stock trend prediction, There are almost 1 and 3 percentage improvement respectively.
Keywords/Search Tags:stock price prediction, behavioral feature, fractal dimension reduction, feature selection, support vector machine
PDF Full Text Request
Related items