Abstract: Chronic Obstructive Pulmonary Disease（COPD） is a progressive disease that makes patients hard to breathe and affects their quality of life. Spirometry is the main way to confirm the diagnosis of COPD.In order to reduce some unnecessary examinations and rebuild the harmonious relationship between doctors and patients. This work adopts the K-L divergence based decision tree to build classification model on the collected clinical dataset and uses this model to screen the patients from the point of view of traditional chinese medical. This experiment compars with three other classic decision trees including ID3, C4.5 and CART across F-Measure, G-Mean, AUROC and AUPRC respectively. The experimental results show that the proposed decision tree obtains the best result among the other classic decision trees and the corresponding values of F-Measure, G-Mean, AUROC and AUPRC are 0.92, 0.894, 0.907 and 0.9. Finally, the proposed decision tree is applied to clinic of TCM. The AUROC of the proposed method reaches 0.823, which achieves expected effect.

Key words: decision tree, K-L divergence;imbalanced datasets, Chronic Obstructive Pulmonary Disease（COPD）, Traditional Chinese medicine

摘要： 慢性阻塞性肺病主要表现为呼吸困难，严重影响了患者的生存质量。肺活量测定法是目前的主要诊断方法。为了构建和谐医患关系，减少过度检查，从中医诊断的角度，根据已收集的病例资料，利用基于KL距离的决策树建立诊断模型，可实现对患者的初步筛查。实验以F-Measure、G-Mean、ROC曲线下面积以及精度召回率曲线下面积作为评价指标，将提出的决策树分别与ID3、C4.5以及CART比较。结果表明，提出的决策树较传统决策树取得了更好的预测效果，对应的评价指标分别达到了0.92、0.894、0.907以及0.9。最后，当应用于临床时，以临床上常用的AUROC作为评价指标，提出的决策树模型达到了0.823，取得了预期效果。

关键词: 决策树, KL距离, 非平衡数据集, 慢性阻塞性肺病, 中医