【作者】 周强；

【导师】 欧阳一鸣；

【作者基本信息】 合肥工业大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 数据库中知识发现(Knowledge Discovery in Database,KDD)结合了数据库技术、人工智能技术以及其它专业领域知识,成为近年来计算机学科研究的热点。目前,KDD研究涵盖了多个方面,在时间序列规则、关联规则、分类规则、聚类规则的发现工作中,取得了较好的效果;在诸如:联机处理分析(OLAP)、数据仓库(DW)等领域的实践工作中,KDD同样得到了广泛应用;此外,随着网络技术的飞速发展,基于WEB的KDD研究也越来越为人们重视。 本文主要研究内容是针对时间序列数据进行分析挖掘,获得内在规律性,并将其作用于金融时序交易应用之中。 在金融领域中,存在着大量数据,由于数据量过于庞大,传统处理方法难于发现其中蕴含的知识,迫切需要新知识、新技术来解决这个问题。KDD技术在金融领域应用,主要集中在客户关系分析与管理方面,对交易数据的挖掘还不多见。而实际工作要求有一种工具可以对交易数据进行分析,发现其内在规律性,从而对交易的性质和发展趋势作出判断。 为此,本文研究了KDD在金融时序数据挖掘中的应用,探索一个合适的模式挖掘方法,设计一个包含挖掘交易模式、分析交易性质并可以预测交易发展趋势的试验系统,以期对KDD在金融领域中应用起到一定的推动作用。 本文主要研究成果如下: 首先,提出一个应用系统框架,该框架结合金融领域知识,完成数据预处理、模式生成与评估,以及数据分析预测等功能。使得人们对时序交易数据的内在规律和特点有更深刻的认识。 其次,针对金融领域中交易数据的时间相关性,结合统计学理论,对时序数据挖掘中的C—均值算法作出了一定改进,使之可以自动地完成模式发现工作。 再次,在时序数据预测时,结合模式匹配,得出小样本空间,以基于小样本的回归预测方法代替数据曲线趋势判断方法,以期得到更佳的预测结果。更多还原

【Abstract】 The Knowledge Discovery in Database combined with database technique, the artificial intelligence and other professional domain knowledges, is becoming a hotspot in computer science in recent years. Currently, the KDD research has related to several aspects such as: time sequence rules, association rules, classification rules and clusters rules. All aspects are obtained good results. In practical works, such as online processing analysis, the data warehouse, KDD got the extensive application. With the rapid development of network technique, people attach more and more important to the research of KDD in Web.This disquisition focus on the research of KDD in time sequence data, acquires the inside regulations and use for financial domain.There are a great deal of data in financial domain. The data quantity is so huge that we can not find knowledges by tradition methods. It’s need new knowledge and technology to resolve this problem. In financial, KDD is mainly used to analysis the custom relationship management. There hasn’t many KDD method to be used in transaction data. In the actual work, it request a kind of tool to analysis transaction data, discover it’s inside regulations, thus to judge these data’s quality and tendency.This disquisition aims at the KDD application in financial sequent data, discover a fit pattern, and design a testing system to predict the data tendence. In expectation, it can rise a certain impetus in financial realm.The followings are results of our research:First, we proposed an applying system frame that can preprocess data, find and evaluate patterns, analysis and estimate data with financial domain knowledges. It helps people to make deeper understanding for the inside regulations and characteristics in data.Second, aiming at the financial time sequence data, we proposed a method to improve C-mean algorithm. So, it can find patterns automatically.Third, in the phase of forecasting, according with the pattern matching, we can get small sample fields. Instead of judgement which based on curve trend, we can get better estimate result by use the regression function in that small sample field.更多还原