基于混淆矩阵的机器学习分类评价指标研究及Python实践

Research on Performance Measure Indicators of Machine Learning Classification Based on Confusion Matrix and Corresponding Python Practice

摘要: 基于混淆矩阵的机器学习分类指标体系是在衡量各个分类器的分类效果中最常用的，本文对这些指标的计算原理进行了全面的整理，其中，对二分类的G-mean值和Matthews相关系数做了三分类及更多分类上的推广定义，并利用UCI的机器学习数据的Vehicle Silhouettes (汽车轮廓)数据集进行了基于Sklearn的相应python实验，给出了相应python代码以及运行结果等。特别地，对于推广到多分类的两个指标定义了相应的python函数，并进行了相应实验和验证。本文为最常用的机器学习分类评价指标的选取提供理论和python实践及相应代码的参考，为广大学者选用指标提供了依据。

Abstract: The machine learning classification performance measure system based on confusion matrix is the most commonly used in measuring the classification effect of each classifier, and the calculation principle of these performance measures are comprehensively listed in this paper. Among them, the G-mean value and Matthews correlation coefficient of two classifications are generalized and de-fined in three or more classification problems. Moreover, using the Vehicle Silhouettes dataset in UCI, the corresponding python experiments are implemented in the bases of Sklearn, and corresponding python codes and running results are given. In particular, the corresponding python functions are defined for the two generalized measures, and corresponding experiments and validations are also implemented. This paper provides a theoretical basis and python practice and corresponding codes for the selection of the most commonly used machine learning classification performance measures for the majority of scholars to select appropriate performance measures.