Analysis And Research On Temporal Association Rules Of Financial Time Series Based On Information Entropy

With the rapid development of each industry of present society,the types and quantities of data explode quickly which is not apparent.And jumbo treasure is hided in such giant and complex data.To mine this treasure,people's requirement can not be got by traditional data mining and knowledge discovery technology.The most key branch of data mining technology is time series association rule,and the mysterious stock trends and rules can be found in historical data step by step because of this technology.Information entropy is an important part of rough set theory,information entropy attribute reduction is the emphasis of information entropy theory research.The complex financial time series can be solved,and the redundant noise can be removed by a useful information entropy data mining algorithm.What is more,the useful and believable association rules can be found by it.From the angle of theory and application,the dissertation researches the financial time series association rule based on information entropy,and the main work are as follows:At the beginning,the concept,types and basic principle of data mining are studied,the information entropy association rule of data mining is deepley researched.On basis of present time series data mining,one of the research guide of this dissertation is found,which is called information entropy association rule.In the second part,a kind of fuzzy c-means clustering algorithm based on adaptive genetic algorithm is proposed.In this algorithm,fuzzy rough set theory and improved adaptive algorithm are combined together,and the proposed AGA-FCM can solve well the problem of information flow loss in the discretization process.The experiment shows that the problem of the local optima of genetic algorithm is overcame by the new mutation operator and crossover formula.And the AGA-FCM clustering algorithm has better clustering effect which is combined with the FCM and GA-FCM clustering algorithm.Last but not least,the time series rules analysis model based on information entropy is proposed in this dissertation,which is applied to the stock predict.And a new financial time series stock prediction model is presented on this foundation.The useful association rules are mined from complex and huge stock technology index data,the redudanct noise of initial data is removed by this model.On contract with traditional neural network prediction model,the burden of neural network machine learning is greatly reduced,the relation and accuracy of import data is improved,and the problems of low training speed of neural network forecast and large memory overhead are solved.Through simulation experiment,it is proved that this financial time series predicting model compared with other models has high predicting accuracy to short time stock closing price.