基于偏最小二乘法的亚健康状态的判别

An analysis of subhealth status based on partial least squares

目的:采用偏最小二乘法探讨亚健康状态的判识模型。方法:采用偏最小二乘法进行建模、预测,对亚健康状态进行判别,统计判别的准确率;在逐步回归变量筛选后再次进行预测,观察判别准确率的变化情况。结果:基于偏最小二乘法建立的亚健康判识模型对亚健康状态的预测准确率为89.47%,经变量筛选后,预测准确率提高至92.10%。结论:偏最小二乘法在亚健康状态的模型判别中具有较高的准确性,在亚健康建模的研究中有一定的参考价值。基于变量筛选后预测准确率的变化,从量表优化的角度来看,偏最小二乘法也可以为变量的精简提供一定的依据。

Objective：To explore the discrimination model of subhealth with statistical method of partial least squares（PLS）.Methods：This study was based on the Subhealth State Rating Scale（SHSRS）.A total of 88 subhealth subjects（scoring less than 85 in SHSRS）and 64 healthy people（scoring over 85 in SHSRS）were enrolled randomly.Information regarding the clinical symptoms was screened by stepwise regression as independent variables.Mathematical models were established by leave-one-out in PLS program for subhealth recognition before and after stepwise regression respectively.Accuracy rates were observed and compared by using the Visual Basic 6.0.Results：The practical accuracy rate of PLS models in subhealth recognition was 89.47%,and increased to 92.10% after stepwise regression for variables.Conclusion：PLS has certain reference value in establishing subhealth discrimination models.It can also play an important part in item selection of the scale.