Data Mining Technology And Its Applications In Securities

In this doctoral dissertation, we discuss the theory and technology of data mining and their applications in the securities fields for the following four aspects.First, we research on the clustering method of customer analysis. Through mining customer historical transaction, we propose and realize three logical functions of customer division model based on the Customer Relations Management (CRM): customer automatically clustering mechanism, customer grouping mechanism in the same clustering class, and dynamic renewal mechanism of the model. In realization of customer automatically clustering mechanism, we establish the clustering indicator set, including four indicators: the capital scale, the commission, the interest balance income, the trading frequency. Through statistical correlation analysis, we define the weight of every indicator using the clustering similarity formula, and put forward a performance indicator to evaluate the effect of customer clustering by analyzing the result of customer clustering. In the same class, we propose and realize customer grouping method according to the customer operation style by using dynamic Bayes model.Secondly, we propose and apply the method of constructing customer behavior model of Internet securities dealing system by mining association rules and sequential patterns from historical trading data warehouse. In terms of the transition matrix of Markov chains constructed through customer behavior model, we realize the push algorithm of stock quotations in the Internet securities dealing system. The algorithm effectively enhances real-time response performance of stock quotations pushing system.Thirdly, according to the importance of Churn Analysis of securities business and the RFM model of Customer Relationship Management theory, we propose and apply the RFM-ROI model, and demonstrate it by quantitative analysis approach based on information gain theory. In the light of the RFM-ROI model, we construct a decision tree from mining historical trading data warehouse and extract the rules of Churn management. Furthermore, we put forward optimal stopping threshold method, which is selected as a tree pruning strategy.Finally, we propose a new algorithm for mining some typical patterns in the stock time series based on RBF neural network, which is meaningful reference for the investors and could decrease the subjectivity of the investors. We transform the time series into a same standard through inserting time points and normalizing. The results show that RBF neural network can well classify technical patterns after preprocessing the input vector.