一种应用于智能分诊的改进朴素贝叶斯方法

An improved naive bayes method applied to intelligent diagnosis

针对朴素贝叶斯分类方法(naive bayesian model,NBM)在应用于门诊智能分诊时,无法有效区分不同类型的症状涉及的疾病学科范围不同问题,提出了一种朴素贝叶斯分类方法的改进算法,引入IDF因子,为不同的症状类型提供相应的权重。首先,基于权威医疗文献,收集整理诊断学相关的语料作为训练数据集,然后,基于朴素贝叶斯分类方法计算先验概率、类条件概率,训练生成不同症状的IDF因子,最后,在进行分类判断时对不同的症状组合引入IDF因子,平滑不同类型症状的重要程度。在智能分诊准确性对比实验中,改进后的算法召回率提升约11%,明显高于朴素贝叶斯分类方法。

Focusing on the issue that the naive Bayes model(NBM)in outpatient intelligent diagnosis,it is not effective to distinguish between different types of symptoms involved in a different range of subjects.An improved algorithm for the naive Bayes method is proposed,Introducing IDF factor,Provide different weights for different symptom types.First of all,based on authoritative medical literature,Collected and sorted the related corpus of diagnostics as a training data set,Then,based on the naive Bayes method,the priori probability and the class conditional probability are calculated,Trained the IDF factors for different symptoms,Finally,IDF factor is introduced to different combination of symptoms in classification judgment,to smoothed the different types of symptoms.In the accuracy comparison experiment of intelligent diagnosis,the recall rate of the improved algorithm is up about 11%,obviously higher than the naive Bayes method.