基于一致性预测器的中医证素组合诊断模型

Conformal Predictor Based Syndrome Differentiation for Traditional Chinese Chronic Fatigue Diagnosis

构建中医证素组合智能诊断模型需要特殊的域预测分类器而非传统的点预测分类器.引入一致性预测器(conformal predictor,CP),以算法随机性水平值为证素的重要性度量,以算法风险水平为阈值进行域预测输出,以中医慢性疲劳样本集为研究对象,随机森林(random forest,RF)等传统机器学习算法被嵌入到CP框架中计算样本奇异值.实验结果表明,CP-RF模型不仅拟合率比其他域预测分类器高,还对阈值具有很好的鲁棒性,克服了阈值对预测域的波动性,解决了中医多证素组合诊断关键的技术难题之一,同时CP-RF模型的预测域错误率能够被算法风险水平阈值所校准,表明其阈值具有明确的统计意义和可解释性,能够被中医医生所接受.

Syndrome differentiation in traditional Chinese medicine(TCM)which identifies the combination of some selected syndrome factors as the diagnosis for the patient falls into the region prediction rather than point prediction.In this study,conformal predictor(CP)is introduced in the literature of syndrome differentiation diagnosis which provides algorithms randomness level as the importance of syndrome factor and applies the significance level to serve as the threshold.The study aims to the chronic fatigue(CF) dataset in TCM,for which many novel machine learning algorithms,such as random forest(RF),have been plugged into the framework of CP to compute the nonconformity score of the example.The experimental results show that CP-RF highlights not only significantly high matching ratio beyond other region classifiers but robust to threshold value as well.The latter overcomes the fluctuation of region prediction derives from different threshold values and solves one of the most critical challenges in TCM.Moreover,the error rate of CP-RF is hedged by the significance level,which illustrates statistically interpretability and is easy to acceptance by TCM practitioners.