基于各类支持度阈值独立挖掘的关联改进算法

关联分类及较多的改进算法很难同时既具有较高的整体准确率又有较好的小类分类性能。针对此问题,提出了一种基于类支持度阈值独立挖掘的关联分类改进算法—ACCS。ACCS算法的主要特点是:(1)根据训练集中各类数量大小给出每个类类支持度阈值的设定方法,并基于各类的类支持度阈值独立挖掘该类的关联分类规则,尽量使小类生成更多高置信度的规则;(2)采用类支持度对置信度相同的规则排序,提高小类规则的优先级;(3)用综合考虑置信度和提升度的新的规则度量预测未知实例。在多个数据集上的实验结果表明,相比多种关联分类改进算法,ACCS算法有更高的整体分类准确率,且在不平衡数据上也能取得较好的小类分类性能。

An associative classification algorithm based on variousclass-support thresholds and independent mining rules

Associative classification algorithm and its existing improved algorithms cannot achieve both high overall accuracy and good minority class classification. To solve this problem, we propose an improved associative classification algorithm based on various class-support thresholds (ACCS)independent mining rules. Its main featuresare: (1) ACCS sets the support threshold of each class according to the class size in the training data, and extracts the associative classification rule of each class separately based on the class-support threshold in order to get higher confidence rules of minority classes; (2) ACCS uses the class-support threshold to rank the rules with the same confidence for increasing the priority of the minority classes; (3) ACCS combines confidence and lift degrees together to predict unknown instances. The experimental results on multiple datasets show that ACCS can achieve higher overall classification accuracy than the existing associative algorithms, and can also get good minority class classification performance in imbalanced data.