文档介绍：
基于蚁群优化的改进决策树算法研究
张忠坤,周亚建**
(北京邮电大学信息安全中心,北京 100876)
5
10
15
20
25
30
35
40
摘要: 本文提出了一种新的基于蚁群优化的改进决策树算法(ACODT, Ant Colony
Optimization Decision Tree),该算法充分利用了蚁群算法的信息素反馈和启发搜索特征来改
进决策树的生成过程。选取 15 个 UCI 数据集进行实验并与 和 CART 算法进行比较。
实验结果表明,改进算法适用于连续属性和离散属性,能够有效提高决策树预测精度,降低
决策树规模,增强可理解性。
关键词:决策树;蚁群算法;信息素;启发信息;
中图分类号:
Improving decision trees based on an ant colony algorithm
ZHANG Zhongkun, ZHOU Yajian
(Information Security Center,Beijing University of Posts munications,Beijing
100876,China)
Abstract: In this paper we propose a new decision tree algorithm based on an ant colony
algorithm,which makes full use of the pheromone feedback and heuristic information in the ant
colony proposed algorithm pared against two well-known decision tree
algorithm, and CART,in 15 data experimental results show that the proposed
decision tree algorithm is suitable for boh continuous attributes and discrete can
effectively improve the generated decision tree's prediction accuracy,reduce its size and enhance
its intelligibility.
Key words: decision tree; ant colony algorithm; pheromone; heuristic information;
0 引言
决策树算法是机器学习领域具有代表性的算法之一,具有分类速度快、精度高、生成过
程简单以及产生的规则易于理解等优点,已经被成功地应用到计算机辅助医疗诊断、数据挖
掘、风险评估等领域。但是,决策树算法也存在着对噪声敏感、在测试属性选择时倾向于选
择取值较多的属性等缺陷。
蚁群算法是一种模拟真实蚁群的觅食行为而形成的仿生算法,具有分布计算、信息正反
馈和启发式搜索等特征,易于与其它方法结合,形成优化算法[1][2]