基于随机森林的医疗行业供应商的等级评估模型

发布时间：2018-05-22 19:09

本文选题：随机森林 + 指标　；参考：《安徽大学》2017年硕士论文

【摘要】：目前,传统的市场经济环境正在发生着巨大的演变。企业现在所面临的不仅仅是以往常见的企业间的竞争,而是转变为企业在供应链上的资源争夺。而随机森林算法成为现在在科学操作方面崭新的一种使用方法。它主要被用来发掘大量数据中隐藏的那些可以使用的,能够在现实中操作的信息。而本文就选择随机森林这种算法作为实验的依据。首先,对论文的研究背景,意义和方法进行了简单的描述。而目前来说,大多数企业在做供应商评估时都没有一套科学的体系,都是评估人员凭借着自己的经验来进行。这种方法存在较强的主观性,而建立合理的评估体系对企业降低成本,减少风险有着很大的好处。但是好的体系的建立需要选出那些具有代表性的指标。通过对国内外的文献进行研究,选取指标时要能够严格的契合研究的目的;指标体系的构建要满足:完善性,合理性,易操作性。在第二章中对随机森林算法模型进行了详细的阐述。随机森林是树型分类组合器中的一种,对样本数据的处理采用Bagging和随机选择特征的方式进行。而在使用Bagging的方法进行抽样时,会有一部分数据不会被抽中,这部分数据就可以用来估计模型的泛化误差。同时,通过实验证明得出,随机森林模型的泛化误差在树的数目达到一定值时,其收敛于一个有限值,所以利用这个原理可以确定森林中树的数目。而根据本文研究的目的,选择了 22个指标,由于随机森林模型可以计算指标体系的重要性,借助于R软件,通过实验可以得到最终的指标体系。那么在最后一个章节就是根据得到的指标体系来建立随机森林模型,并且通过随机森林验证了其对噪声具有很好的免疫能力。本文这种对供应商通过建立模型对其进行评估的方法,是值得进行深入推广研究的。实践和理论的相互结合,证明了随机森林有着很好的性能。但是,本文在研究中也有一些问题值得进一步的思考,如选择的数据量不是很大,可能会存在偏差,而且对存在的离群点没有做任何的处理,同时指标体系在筛选时并没有进行详细的解释。
[Abstract]:At present, the traditional market economy environment is undergoing a tremendous evolution. What enterprises are facing now is not only the competition among enterprises, but also the competition for resources in supply chain. The stochastic forest algorithm has become a new method in scientific operation. It is mainly used to extract information hidden in large amounts of data that can be used and can be manipulated in reality. In this paper, random forest algorithm is chosen as the experimental basis. First of all, the research background, significance and methods of the paper are briefly described. At present, most enterprises do not have a scientific system in supplier evaluation. This method has strong subjectivity, and the establishment of a reasonable evaluation system has great benefits to reduce the cost and reduce the risk. But the establishment of a good system requires the selection of representative indicators. Through the study of the domestic and foreign literature, we should be able to strictly fit the purpose of the research when selecting the index; the construction of the index system should be satisfied: perfect, reasonable, easy to operate. In the second chapter, the stochastic forest algorithm model is described in detail. Random forest is a kind of tree type classifier. The processing of sample data is carried out by Bagging and random selection. When sampling with Bagging, some of the data will not be extracted, which can be used to estimate the generalization error of the model. At the same time, it is proved by experiments that the generalization error of the stochastic forest model converges to a limited value when the number of trees reaches a certain value, so the number of trees in the forest can be determined by using this principle. According to the purpose of this paper, 22 indexes are selected. Because the stochastic forest model can calculate the importance of the index system, by means of R software, the final index system can be obtained through experiments. In the last chapter, the stochastic forest model is established according to the obtained index system, and its immunity to noise is proved by random forest. In this paper, the method of evaluating suppliers by establishing models is worthy of further study. The combination of practice and theory proves that stochastic forest has good performance. However, there are some problems worth further thinking in this paper. For example, the amount of data selected is not very large, there may be deviation, and there is no treatment of the outliers. At the same time, the indicator system in the screening and did not carry out a detailed explanation.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;TP18

【参考文献】