customer behavior:客户行为检测决策树数据分析Detecting the change of customer behavior based on decision tree analysis [9]
论文作者:留学生论文论文属性:硕士毕业论文 thesis登出时间:2011-01-14编辑:anterran点击率:25053
论文字数:9758论文编号:org201101141122036887语种:英语 English地区:韩国价格:免费论文
关键词:data miningdecision treechange analysisInternet shopping mall
Vol. 22, No. 4do not know how much change has occurred. Thereforeadditional logical judgement is required to concludewhether the degree of change is significant or not. For thispurpose, we adapt the unexpectedness concept from thestudy of Padmanabhan and Tuzhilin (1999). They defineunexpectedness using the exception rule concept (Suzuki,1997; Hussain et al., 2000) as follows.Definition 5: Unexpectedness If a rule from a decisiontree A ) B is unexpected with respect to the belief X ) Y,then the following must hold.(1) B and Y¼False.(2) The rule X, A ) B holds.Anew measure for the degree of change of an unexpectedconsequent change is defined using Definition 5. Tomeasure the degree of unexpected consequent change, rtiis assumed to be a belief or existing knowledge. Everyunexpected consequent change satisfies condition (1) ofDefinition 5 because of Definition 2. Furthermore thesupport value of the conjunction rule should be evaluatedto check whether condition (2) of Definition 5 holds or not.For example, a conjunction rule of the above example is asfollows.ri\j : Income=High, Age=High, Preference=Price) Sales amount =LowIf the above conjunction rule, ri\j , is statistically large (i.e.has a large support value), then we can conclude that rtþkj isan unexpected consequent change with respect to rtibycondition (2) of Definition 5. Therefore, the support valueof the conjunction rule can be regarded as the degree ofchange for unexpected consequent change. But the twoconditions of Definition 5 are not sufficient. If the supportvalue of the conjunction rule is relatively small bycomparison with the support value of rtþkj , then we cannotconclude that rtþkj is a significant unexpected consequentchange with respect to rti. An additional condition thatshould be included is that the support value of ri\j shouldbe large enough to represent rtþkj . Therefore, the degree ofchange for unexpected consequent change should becomposed of the support value of rtþkj and ri\j . Now weprovide the following measure for the degree of unexpectedconsequent change.aij ¼suptþkðri\jÞsuptþkðrjÞIn the case of an emerging pattern, it is simpler toevaluate the significance level than the case of unexpectedchange. The growth rate or rate of decrease are used as themeasure for this type of change. To evaluate the degreeof change for the added and perished rule cases, thesupport value of these rules and the maximum similarityvalue are used. As mentioned before, the maximumsimilarity value of a rule represents the degree of similarityof the most similar rule to the other rule set. If there isa situation that the support values of two added rules arethe same, we naturally place more importance on the rulewhich has less maximum similarity value. Such a rule givesmore significance than the other rule. The measure of thedegree of change, aij, is summarized as follows. Based onthe value of aij, we can rank the changed rules in each typeof change.aij ¼suptþkðriÞ suptðriÞsuptðriÞemerging pattern casesuptþkðri\jÞsuptþkðrjÞunexpected change caseð1 siÞsuptðriÞ perished rule caseð1 sjÞsuptþkðrjÞ added rule case8>>>>>>>>><>>>>>>>>>:5. Experiments and applicationsA case study has been conducted to evaluate how well theprocedure performs it
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。