With development of computer network and appearance of big data,researchers put more and more focuses on forecast of stock transaction.How to take advantage of big data produced in stock transaction,get stable factors which can affect the future of stocks,and filter the factors which can disturb the forecast is very important in application.Firstly,this thesis introduces current research in the world,analyzes the solved problems in the researches,summarizes the existing problems n stock forecasting,and presents the focus in this thesis.Secondly,the thesis descripts some basic data mining concepts.The clustering algorithm with k-means and DBSCAN is detailed.Meanwhile,the classification algorithm with decision tree and neural network is introduced.These algorithms provide necessary theory for the research of this thesis.Thirdly,aimed to get short-term profit,this thesis analyzes the feature of rising stock,and a new intelligent stock selection based on multi-level clustering is provided.It analyzed the stock transaction records,and classified the data into rising and unrising categories.Then it clusters the rising transaction data into different groups with different levels.Combined with the oridinal unrising category,the data set is formed.Based on this data set,decision tree algorithm is apoted,and the classifying model is achieved.Then the result is collected,and the final modle is achieved which can get only rising and unrising data.Fourthly,the intelligent stock-selection algorithm is used.And an experimental policy,which includes target definition,data preparation,clustering,lassification and analyses of experimental result,is present with the stock transcation data of 899 stocks from 2006 to 2010,665621 records in total.The definition of rising data is present,and the division procedure of transaction data is introduced detailly.Then it uses K-means as its' clustering algorithm,and uses C50 as its' classification algorithm.The experimental result is achieved.It shows that the new algorithm get better performance compared with other classification algorithms.Finally,a prototype of intelligent stock-selection system is realized with new algorithm and Visual Basic.It presents design of system and modules.Using transaction data,this system successfully forecast some stocks from Apr 24 th,2015 to May 22,2015,and get satisfying result. |