The Securities Situation Assessment System Based On Data Mining

This paper was one of the core components of the intelligent search system on securities. It focused on technique analyzing, namely using data mining on trading data to set up an auto deciding system for securities investment. Firstly, the paper was detailed on the system architecture. Then we presented some applications with data mining. Finally, the speech recognition interface was introduced.We adopted the situation assessment theory of data fusion on securities field, and put forward a new concept: Securities Situation Assessment. Because of the high complexity of securities market, it was impossible to make an accurate prediction through a single approach. To make use of advantage of multi predicating methods, the paper used data fusion technique to organize all kinds of data mining means. Because of hierarchical requirement of situation assessment, we realized the securities situation assessment system (SSAS), and constructed a securities data mining method ontology database. The system was designed as an open platform and new analyzing techniques could easily be appended to the system.The paper clustered the closing price, price - earnings ratio, price - sales ratio, price - net asset value ratio and circulated capitalization with self organizing feature mapping network (SOM), as set up a block definition method based on finance index. Ran wavelet to the time series of a stock, gained energy and energy distribution on every level wavelet decomposing, got the information measure coefficient, then clustered with SOM, it create a new block compartmentalizing method.The paper used Apriori algorithm mining association rules among single stock, then carried through association rules mining based on clustering. The paper also analyzed the stock tendency in every working day and the mined association relation among them, the result indicated that the time anomalies did exist in the Chinese securities market.The paper used Markov chain to predict the stock price tendency. We utilized the wavelet coefficient as the observed value, following the digital speech recognition thinking, predicted the stock price tendency with the hidden Markov Model (HMM), and got an improved effect. Also, we applied the back propagation neural network (BP) to predict the stock price. The results showed the flexibility BP and conjugate BP had a better predication effect.