结合属性关联度的决策树算法研究及应用

发布时间：2019-03-17 17:16

【摘要】：在新的世纪,挑战与机遇并存,可以说利用并驾驭所产生出的海量数据关系着各个行业未来的发展。在大数据领域的深入探索可以更宏观的分析数据模式,发现潜在规律并对未来趋势进行合理预测,从而可以洞察到更为深刻、有效、全面的信息。数据挖掘相关算法的研究,可以说是兼具科学研究价值与实用价值。本课题在经典的决策树C4.5算法基础上,利用Apriori关联规则算法将数据源属性间的关联度结合到后期决策树的计算中。传统的C4.5算法对选择分裂属性时只是考虑待测属性与类属性之间的相关性,忽视了非类属性间的关联程度,而这种关联程度决定了属性间冗余度的大小。为了能够降低冗余带来的影响,本文运用信息增益的思想对待测属性和其他非类属性进行度量,并加入到原始的算法中,从而生成更加可靠的分裂属性。另外在决策树模型的构建过程中,针对属性涵盖信息量不足的情况,论文同时也采用了关联规则Apriori算法生成一系列强规则,并根据提出的新属性选择标准,从这些强规则中继续筛选出新的属性并加入到原始属性集合中,达到扩充信息量的目的,进而使C4.5算法预测的准确率得到提升。一个样例中包含的信息往往是多样而丰富的。利用传统的决策树算法,我们可以知道"属性-分类"的关联程度,但属性间的关联程度是一种对数据集的横向分析,通过分析两个属性间的关系可以使我们的分析框架更为立体,结果更具有可用性。最后,本文将该方法应用到了实例中,利用历史数据找出影响健身场馆顾客入会的主次因素,结合相关属性建立模型并进行预测,发现那些有兴趣并且有较高价值的客户群体,进一步说明算法在实际场景中的实用价值。
[Abstract]:In the new century, challenges and opportunities coexist, it can be said that the use and control of the mass of data is related to the future development of various industries. The in-depth exploration in big data's field can analyze the data model more macroscopically, discover the potential law and predict the future trend reasonably, so as to gain insight into more profound, effective and comprehensive information. The research of data mining related algorithms can be said to have both scientific research value and practical value. In this paper, based on the classical decision tree C4.5 algorithm, the Apriori association rule algorithm is used to combine the correlation degree among the attributes of data source into the calculation of the later decision tree. The traditional C4.5 algorithm only considers the correlation between the attributes to be tested and the class attributes when selecting the split attributes, and ignores the degree of association between the non-class attributes, which determines the degree of redundancy between the attributes. In order to reduce the influence of redundancy, this paper uses the idea of information gain to measure the measured attributes and other non-class attributes, and adds them to the original algorithm to generate more reliable splitting attributes. In addition, in the process of constructing the decision tree model, in view of the lack of information, the paper also uses the association rule Apriori algorithm to generate a series of strong rules, and according to the proposed new attribute selection criteria, the paper also uses the association rules Apriori algorithm to generate a series of strong rules. From these strong rules, new attributes are screened out and added to the original set of attributes to expand the amount of information, so that the prediction accuracy of C4.5 algorithm can be improved. The information contained in a sample is often diverse and rich. By using the traditional decision tree algorithm, we can know the degree of association of "attribute-classification", but the degree of correlation between attributes is a kind of horizontal analysis of data set. By analyzing the relationship between the two attributes, we can make our analysis framework more stereoscopic and the results more usable. Finally, this paper applies this method to practical examples, using historical data to find out the primary and secondary factors that affect the entrance of gymnasium customers, establish the model and forecast the relevant attributes, and find those interested and valuable customer groups. The practical value of the algorithm in the actual scene is further explained.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13

【参考文献】