基于改进Ant-miner算法的分类规则挖掘  

Classification Rule Mining Based on Improved Ant-miner Algorithm

作　　者：肖菁[1] 梁燕辉[2] 

机构地区：[1]华南师范大学计算机学院,广州510631 [2]中山大学计算机科学系,广州510006

出　　处：《计算机工程》2012年第17期162-165,共4页Computer Engineering

基　　金：教育部博士点新教师基金资助项目(20090171120003)

摘　　要：为提高基于传统Ant-miner算法分类规则的预测准确性,提出一种基于改进Ant-miner的分类规则挖掘算法。利用样例在总样本中的密度及比例构造启发式函数,以避免在多个具有相同概率的选择条件下造成算法偏见。对剪枝规则按变异系数进行单点变异,由此扩大规则的搜索空间,提高规则的预测准确度。在Ant-miner算法的信息素更新公式中加入挥发系数,使其更接近现实蚂蚁的觅食行为,防止算法过早收敛。基于UCI标准数据的实验结果表明,该算法相比传统Ant-miner算法具有更高的预测准确度。In order to improve the classification rule accuracy of the classical Ant-miner algorithm, this paper proposes an improved Ant-miner algorithm for classification rule mining. Heuristic function with sample density and sample proportion is constructed to avoid the bias caused by the same probability in Ant-miner. A pruning strategy with mutation probability is emploied to expand the search space and improve the rule accuracy. An evaporation coefficient in Ant-miner＇s pheromone update formula is added to slow down the convergence rate of the algorithm. Experimental results on UCI datasets show that the proposed algorithm is promising and can obtain higher predication accuracy than the original Ant-miner algorithm.

关 键 词：Ant-miner算法 分类规则挖掘 数据挖掘 蚁群优化 规则修剪策略 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]