Research On Mining Association Rules From Stock Time Series Data

Data mining aims to get previously unknown and potentially useful knowledge from a large amount of data to offer decision support. With the development of the stock market, lots of history exchange data has been stored in database. It attracts more and more attention that how to use these history exchange data to discover the rules of the stock market. Association rules mining is an important problem in data mining research field. The aim of the association rules mining is to extract associations from vast data or objects. According to the associations we can discover the interdependence among objects and infer the property of an object from the property of the other. As a kind of the time series data, the stock one has the special character of his own besides the general characters of the time series data. If we can do some exploring research on the stock time series data via applying advancd data mining technology (such as association rules mining), which is based on the traditional economic & statistical analysis method, and get the potentially valuable knowledge, this research, aparantly, has signality in theory and practice. This paper probes into the above problems, the main contents and research productions include 3 aspects as follows:1. Detailedly analyze the application of all kinds of data mining technologies based on the status quo of the data accumulation in securities business, and presents an architecture model of data mining system in securities by referencing the general data mining system architecture. What's more, we research the basic procedure and function components of the architecture model by the numbers.2. Discuss the classical algorithms for the association rules mining and general methods for time series data analyzing. According to the special propety of the stock time series data, we present an algorithm guided by Meta-Rule, which is basd on the Apriori algorithm, for mining the association rules aiming at describing the interdependent changes of the stock price. Firstly, we get the transaction sets which are suitable for mining rules after preprocess the original stock data by adopting the Sliding-Time-Window technology; Afterward, we detailedly discuss the procedure to construct the association rules by using SQL.3. Discuss the data mining model based on Rough Set and data reduct algorithms. A new algorithm, which is based on Rough Set, aiming to discover association rules in stock time series data is presented. The algorithm includes 3 steps: data preprocessing, data reduct and rules extracting. The association rules attainted by using the new algorithm can preindicate the future trend of the stock time series data.