【作者】 卢锦；

【导师】 甘岚； 卢春；

【作者基本信息】 华东交通大学 ， 计算机技术， 2010， 硕士

【摘要】 关联规则是数据挖掘领域中一个非常重要的研究课题,其本质是揭示出隐藏在大量数据中的对象之间的依赖关系,根据这种依赖关系可以从某一对象的信息来推断相关对象的信息。股票时间序列数据如能运用数据挖掘技术对其进行探索性挖掘出潜在的有价值的模式,在理论研究和实践指导上都具有重要的意义。首先对数据挖掘中的关联规则挖掘算法做了分析,然后着重开展了如下两个方面工作:①将时间序列变换到离散(符号)的事务数据后,利用传统经典Apriori算法,对IT股票间连动规则进行关联规则建模,同时详细阐述了模型设计思想和算法的实现。②利用时间序列相似性搜索方法,查找出与选定股票走势相似的历史股票数据。对时间序列进行平滑规范化处理后,使用欧几里德距离进行聚类,找到频繁时间序列片段,并用时间序列片段模式匹配方法,对频繁时间序列片段进行关联规则分析,直接从数据中学习频繁发生的模式,并用此模式来进行趋势预测。基于以上工作,应用原型系统,以沪深A股中IT板块102个股票2007年3月至2009年7月股票收盘价格作为测试集,对IT股价时间序列中频繁片段模式的关联规则进行挖掘预测,在得到的结果中,发现匹配结果相同的字符串所对应的历史数据具有很大程度的相似性,验证了未来的股市变化将是以往某一阶段历史的重现,论证了该方法简单可行,而且是有效的,比传统关联规则算法更适合股票预测,并且对于一些具有延时性的序列能够得出良好的效果。对挖掘原型系统进行设计,并实现了相应的挖掘算法。更多还原

【Abstract】 Data mining association rules is a very important research topic, which essence is revealed hidden dependencies between the objects in a large number of data set, according to this dependency information from an object to infer information about related objects. With stock time series data using data mining techniques to explore its excavation a potentially valuable pattern, theoretical study and practical guidance is of great significance.First, the data mining association rule mining algorithm was analyzed, and then focus on two aspects of the work carried out as follows:①After transforming source data to the discrete time series (symbol) of transaction data on which traditional classical Apriori algorithm is used, inter-linked shares association rules about the IT is modeled, and design detail of the algorithm model is predicted.②By using time series similarity searching method, historical stock data is found out according to the similarity about trends. Following smooth the time series after normalization, by clustering with Euclidean distance the frequent time series fragments are found, on which time series fragments frequent association rules are analyzed, learning directly from the data pattern of frequent occurrence, and use this model to trend forecasting. Based on the above, application of prototype system to 102 stocks in the IT section of Shanghai and Shenzhen A shares from March 2007 to July 2009 closing stock price as the test set, the time sequence of the IT frequent fragment mining association rules model predicts observing systems in the analysis from a large number of search results matching the results found in the same strings with the historical data corresponding to a large degree of similarity, which shows changes in the future the stock market that will be in the past to reproduce a certain stage of history. That demonstrates the method is simple and effective. Comparing with the traditional association rules algorithm it is more suitable for stock forecasting, and can direct good results for a number of sequences with delay. On the establishment of the corresponding prototype system, the corresponding algorithm has been implemented.更多还原