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 [8]
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:25051
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
cates whether the rule isadded or perished. If si<RMT, then rtiis recognized as aperished rule. If sj<RMT, then the rule rtþkj becomes anadded rule.Unexpected change consists of unexpected conditionsand unexpected consequents. To detect an unexpectedchange, a difference measure is provided as follows.dij ¼‘ijPk2AijxijkAij yij if Aij 6¼ 0; cij ¼1 yij if Aij ¼0; cij ¼18>><>>:As defined in the problem definition section, if conditionalparts are similar but consequent parts are different, thenthis rule is called an unexpected consequent. If the rule rtþkjbecomes an unexpected consequent change with respect torti, the similarity in the conditional part and the difference inthe consequent part should be large. Therefore, if dij>0,then rule rtþkj is an unexpected consequent change withrespect to rti. If dij<0, then rule rtþkj is an unexpectedcondition change with respect to rti. If dij¼0, then the tworules rtiand rtþkj are the same rules or completely differentrules. Additional measures such as ‘ij, yij etc. should beprovided in the case dij¼0. If these values are 1 then we candirectly find that the two rules are the same. We computedifference measures between two rules rtiand rtþkj only inthe case of cij¼1. If attributes of consequent parts betweentwo rules are different, it makes no sense to compare thedegree of difference because these two rules are completelydifferent rules.Finally, we should resolve duplication of type. Forexample, although rtþkj is judged to be an unexpectedchange with regard to rtiby the difference measure, wecannot conclude directly whether it is an unexpectedchange or not, because rtþkj can be an emerging patternwith regard to rtjwhich has the same structure as rtþkj . Inthis case, rtþkj should be classified as an emerging patternand not as an unexpected change. As we cannot concludebased on dij alone whether rtþkj is an unexpected change oran emerging pattern, we provide the following modifieddifference measure.d0ij ¼ dij kijwherekij ¼1 if maxðsi; sjÞ¼10 otherwise The fact that si (or sj) is equal to 1 means that the samerule exists in another rule set. That means that rtþkj is likelyto be classified as an emerging pattern. If d0ij is greater thanthe pre-specified RMT, then the rule rtþkj is concludedto be an unexpected change with respect to rti. Table 1summarizes the value of each measure for each type ofchange.4.4. Evaluating the degree of changeAll the changed rules have to be ranked by the degree ofchange. We will explain the idea of evaluating the degree ofchange for each type of change. First, let us consider anunexpected change. The following example presents whyadditional measures are required.rti: If Income¼High, Age¼High, then Salesamount¼Highrtþkj : If P
Reference¼Price, Age¼High, then Salesamount¼LowIf RMT is set equal to 0.4, then the rule rtþkj becomesan unexpected consequent change with respect to rtiasdij¼0.5. But there are two problems in concluding whetherthis change is significant or not. First, we cannot capturethe change easily because the conditional parts are not thesame. Second, although we can understand the change, we
Expert Systems, September 2005,
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。