使用多支持度的关联规则分类算法

摘要

传统关联分类算法使用单一最小项目支持度挖掘关联规则,导致稀有项关联规则无法被发现,从而影响分类的准确性和实用性.提出一种多支持度关联规则分类算法MS-CBAR(Multiple Supports-Classification Based on Association Rules),将多最小项目支持度模型应用于关联分类,以有效挖掘稀有项.该算法为数据库中的规则项提供了用户可定义的最小项目支持度.MS-CBAR算法使用项的最小项支持度阈值、类的最小类支持度值和规则项的最小支持度值决定分类规则是否频繁.生成分类规则集后,使用最高优先度规则覆盖法基于规则集建立分类器.实验表明,所提算法在包含稀有项目及稀有类的数据集中准确率高于传统关联分类算法及其相关算法,表现更稳定.%Traditional association classification algorithm uses single minimum item support to mining association rules,resulting in rare item association rules hard to find,thus affecting the accuracy of the classifier and practicality.Therefore,we propose a multiple support association rule classification MS-CBAR algorithm (Multiple SupportsClassification Based on Association Rules).Besides,the multi-minimum project support model is applied to association classification to effectively exploit the rare items.This algorithm provided user-definable minimum item support for both rule items and classes in the database.Then,the MS-CBAR algorithm adopted the minimum item support threshold,the minimum class support value of the class and the minimum support value of the rule items to determine whether the classification rules are frequent.Finally,after generating the classification rule set,the top priority rule coverage method was used to build the classifier based on the rule set.Experimental result demonstrates the proposed algorithm is more accurate than traditional association classification algorithms in data sets with rare items and rare classes.And its related algorithms are more stable.