Abstract: The paper focus on the knowledge mining of TCM “five pathogens produced by five organs” in the medical corpus by random forests.Firstly，the intelligent diagnosis classification model of “five pathogens produced by five organs” is established.Secondly，an improved feature importance estimation method based on the random forests is applied to evaluate the importance of features for each class.Empirical study shows that the model used is effective and advantageous，and can be competent for exploration the thorough and meticulous conclusion of the fire，heat，wind，cold，damp and body fluid.The proposed method provides excellent decision-making tool for TCM clinicians.

Key words: medical records, five pathogens produced by five organs, classification, random forests

摘要： 采用随机森林算法对语料库中医“内生五邪”病证知识进行数据挖掘。采用随机森林对五邪病证的临床特征体系构建“内生五邪”智能诊断模型。采用改进的随机森林特征重要性度量方法针对各个类进行特征重要性度量。实验结果验证了该方法的有效性和优越性，能够胜任对疾病过程中所产生的类似于风、寒、湿、燥、火邪等五种病理状态进行深入细致的客观化研究。该研究将为中医临床医生提供一个诊疗决策的优良工具。

关键词: 医案, “内生五邪”, 分类, 随机森林