决策树算法及其在冠心病诊疗中的应用研究

发布时间：2018-03-14 09:35

本文选题：多值属性多类标　切入点：决策树　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来冠心病以其高发病率、高致死率的特点,给国人带来了健康威胁与经济负担。发展迅速的计算机技术为探索蕴含在中医诊疗数据中的疾病防治规律和用药知识提供了技术基础。如何从已有的数据中挖掘到这些规律和知识,以及如何对这些规律和知识加以利用来为冠心病中医诊疗提供辅助决策将是本文要研究的主要内容。决策树算法结果表达直观易理解。为了直观地反映冠心病症状与证候之间的关系,本文以多值属性多类标决策树算法为研究对象,针对该算法在冠心病中医诊疗数据处理过程中出现的问题进行相应的改进,并通过实验证明了改进的有效性。具体研究工作如下:第一,原有的多值属性多类标决策树算法在分裂属性选取的过程中会忽略掉当前属性取空值的数据。冠心病数据具有多缺失值的特点,原算法会大量丢失数据,从而使分类准确率相对较低。针对这个问题,本文对算法属性选择进行了改进,增加了对空值的判断,并将当前属性值为空的数据作为新的子结点,从而保证数据不丢失。第二,若数据中出现过多的空值属性,按照上述方法处理时容易出现过拟合的现象,造成决策树规模过大,模型分类准确率不高的问题。针对这个问题,本文在建树之前引入空值数目阈值,并根据该阈值对数据集进行预处理,排除掉空值数目过多的数据,这样分类准确率下降过快的问题得到解决。第三,在分裂效果评价阶段,本文提出了新的相似度计算公式,使类标集之间的相似度计算更加合理,并且公式中的参数能够反映集合之间的特征,根据这个特征自动调节。第四,实际应用方面,本文设计了一个辅助诊疗系统,系统中应用了上述改进后的算法。根据选定的症状来对中医证候进行预测,得到患者的中医诊断结果供医生参考。另外,系统中分类模型的训练集能够实现动态增加,在系统之上能够扩展出用于不同目的的专题挖掘子模块。
[Abstract]:In recent years, coronary heart disease is characterized by its high morbidity and high mortality. The rapid development of computer technology provides a technical basis for exploring the laws of disease prevention and treatment and the knowledge of drug use contained in the data of TCM diagnosis and treatment. How to excavate from the existing data. To these laws and knowledge, And how to make use of these rules and knowledge to provide assistant decision for the diagnosis and treatment of coronary heart disease will be the main content of this paper. The results of decision tree algorithm are intuitionistic and easy to understand. In order to reflect coronary heart disease intuitively. The relationship between symptoms and syndromes, In this paper, the multi-valued attribute multi-class decision tree algorithm is taken as the research object, and the corresponding improvement of the algorithm in the process of data processing of TCM diagnosis and treatment of coronary heart disease is carried out. The effectiveness of the improvement is proved by experiments. The specific research work is as follows: first, The original multi-valued attribute multi-class decision tree algorithm will ignore the data of the current attribute null value in the process of splitting attribute selection. Coronary heart disease data has the characteristics of multiple missing values, the original algorithm will lose a large number of data. Therefore, the accuracy of classification is relatively low. In order to solve this problem, the algorithm attribute selection is improved, the judgment of null value is added, and the current data with null attribute value is regarded as a new child node. In order to ensure that the data is not lost. Secondly, if there are too many null attributes in the data, it is easy to appear the phenomenon of overfitting according to the above method, resulting in the decision tree scale is too large, and the accuracy of model classification is not high. In this paper, the threshold of the number of null values is introduced before the establishment of the tree, and the data set is preprocessed according to the threshold to exclude the data with too many empty values, so that the problem that the accuracy of classification falls too fast is solved. Thirdly, in the stage of split effect evaluation, In this paper, a new similarity calculation formula is put forward, which makes the similarity calculation between class sets more reasonable, and the parameters in the formula can reflect the characteristics of the set. According to this feature, 4th is automatically adjusted. In this paper, an auxiliary diagnosis and treatment system is designed. The improved algorithm is applied in the system. According to the selected symptoms, the TCM syndromes are predicted, and the results of TCM diagnosis of the patients are provided for the doctor's reference. The training set of classification model in the system can be dynamically increased, and the sub-module of topic mining for different purposes can be extended on the system.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R541.4;TP311.13

【相似文献】