摘要
懒惰式决策树分类是一种非常有效的分类方法。它从概念上为每一个测试实例建立一棵"最优"的决策树。但是,大多数的研究是基于小的数据集合之上。在大的数据集合上,它的分类速度慢、内存消耗大、易被噪声误导等缺点,影响了其分类性能。通过分析懒惰式决策树和普通决策树的分类原则,提出了一种新的决策树分类模型,Semi-LDtree。它生成的决策树的节点,如普通决策树一样,包含单变量分裂,但是叶子节点相当于一个懒惰式决策树分类器。这种分类模型保留了普通决策树良好的可解释性,实验结果表明它提高了分类速度和分类精确度,在某些分类任务上它的分类性能经常性地胜过两者,特别是在大的数据集合上。
Lazy decision tree is a very effective classification method. It conceptually constructs the "best" decision tree for each test instance. However, most studies are done on small databases. In some larger databases, lazy decision tree shows its deficiencies in classification speed, memory consumption, and it is easily induced by noises, which affects its classification performance. On the basis of the analysis on the classification principles of regular decision tree classification model and Lazy decision tree classification model, a new decision tree classification model, Semi-LDtree is proposed. The decision tree nodes generated by the new model contain univariate splits like regular decision trees, but the leaves are egual to Lazy decision tree classifiers. This classification model retains the interpretability of regular decision tree. Experimental results show that this model has higher classification accuracy and faster classification speed, and it frequently outperforms the traditional two models especially in larger databases tested.
出处
《计算机应用与软件》
CSCD
北大核心
2008年第12期229-230,238,共3页
Computer Applications and Software