基于偏最小二乘回归的邮件分类问题研究

Research on Mail Classification Problem Based on Partial Least Squares Regression

摘要: 本文基于最小二乘的主成分回归(PCR)方法对邮件进行分类，进一步使用偏最小二乘回归(PLS)对垃圾邮件识别分类。将PLS与PCR得到的分类准确度进行比较分析，考察PLS分类准确度百分比随分类截点变化的趋势，并得出两种方法下不同k值(主成分个数)对应的ROC曲线图，分析PLS与PCR方法识别和分类垃圾邮件的准确度和稳定性。

Abstract: This paper classifies emails based on the principal component regression (PCR) method of least squares, and further uses partial least squares regression (PLS) to identify and classify spam emails. The classification accuracy obtained by PLS and PCR was compared and analyzed. Then the trend of the percentage of classification accuracy of PLS with the classification cut-off point is examined, and the ROC curve corresponding to different k values (number of principal components) under the two methods is obtained. Finally, this article analyzes the accuracy and stability of PLS and PCR methods to identify and classify spam.