【作者】 王荣；

【导师】 陈纯； 林怀忠；

【作者基本信息】 浙江大学 ， 计算机应用， 2006， 硕士

【摘要】 随着信息采集技术和大容量、低成本存储设备的广泛应用,人们积累的数据呈现爆炸性的增长,但“数据爆炸但知识贫乏”问题的日渐突出,带来了对强有力的数据分析工具的需求,而数据挖掘的出现为这一需要提供了有力的技术支持。数据挖掘(Data Mining),也可以称为数据库中的知识发现,就是从海量数据中提取隐含在其中的、针对某些用户的信息的高级处理过程。 数据分类是数据挖掘的主要任务之一。所谓分类,就是从训练数据中发现同类数据对象的共同属性,建立类的判别模型,用以对新的数据所属类的识别。然而,用于分类的数据往往可能包含有数以百计的属性,其中大部分属性与挖掘任务可能是不相关或是弱相关的,因此,属性选择的好坏对于分类的结果有着很大的影响。属性选择(Attribute selection)就是一个从原有的属性集合中选择一个(相对某种评价准则)最优属性子集的过程。 本文提出了一个基于信息增益和卡方检验的属性选择算法。该方法由两部分组成,首先通过信息增益的计算,对原有的属性进行预处理,留下那些信息增益高(信息量大)的属性;然后利用卡方值的计算,选择那些卡方值大(与目标状态差异大)的属性,作为最后进行挖掘的属性集。 在竞争日益激烈的移动通信行业,如何降低运营成本、提供差异化服务、提高客户的忠诚度和满意度显得尤为重要。在这种情况下,运营商都希望能通过运用客户关系管理(CRM),达到保留有价值客户,挖掘潜在客户,赢得客户忠诚,并最终获得客户长期价值的目的。 在本文的最后,介绍了一下浙江移动的一个客户关系管理系统—离网预测模型的建立过程,并把本文提出的新的属性选择算法应用在这个模型中,取得了不错的效果。经实践证明,此算法在兼顾执行效率的同时,也取得了相当不错的预测准确率。更多还原

【Abstract】 The electronic data gathering devices have been wildly used and the data storage with large volumes have become easier and cheaper, which lead to a data explosion. However, in fact, we are drowning in data but starving for knowledge, which pulls the demands of powerful data analysis tools. The emergence of data mining provides strong technical support for the urgent need. Data Mining, also known as KDD (Knowledge Discovery in Database), is an advanced process, in which we can pick up hidden information to certain users from large volumes of data.Classification is a main task of the data mining. Classification is a process of finding the common features in the same type of data object from training datasets, building model, in order to ensure which class the data belongs to. However, the data used for classification contains hundreds of attributes, most of which are irrelevant or weakly relevant to the classification, so feature selection plays an important role in classification. Attribute selection is a process of selection a best subset (according to some criteria) from the dataset.This paper proposes a new future selection algorithm based on information gain and chi-square test. This algorithm is composed of two parts. First, we select futures by information gain. We get the attribute with higher value into next step. Second, we generate the final attributes set for data mining by filtering the attributes with lower value of chi-square.In the growing competitive telecom industry, how to reduce operational cost, to offer differentiated services, and to improve customer loyalty and satisfaction is substantially important. Under this circumstance, the application of customer relationship management could help telecom service operators to retain valuable customers, explore potential customers, win customer loyalty and finally achieve long-term customer value.At the end of this paper, we introduce a Customer Relationship Manager system of Zhejiang mobile named Churning Prediction Model. The new feature selection algorithm proposed by this paper was applied in this system, which shows that the algorithm is efficient, also has a high classification accuracy rate.更多还原