Coefficients: (1 not defined because of singularities)

最新推荐文章于 2022-11-07 14:49:31 发布

Data+Science+Insight

最新推荐文章于 2022-11-07 14:49:31 发布

阅读量1.8k

点赞数 3

文章标签： r语言人工智能机器学习数据挖掘深度学习

本文链接：https://blog.csdn.net/zhongkeyuanchongqing/article/details/120626107

版权

本文探讨了如何处理数据中x1和x2高度相关导致的共线性问题，通过相关矩阵分析发现相关性，移除强相关变量x2，然后重新构建了 glm 模型。模型总结显示部分系数未定义，原因在于奇异矩阵。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

Coefficients: (1 not defined because of singularities)

问题：

解决：

完整错误：

问题：

#模型的特征存在共线性问题，有些特征强相关；

#define data
df <- data.frame(y = c(0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1),
                 x1 = c(3, 3, 4, 4, 3, 2, 5, 8, 9, 9, 9, 8, 9, 9, 9),
                 x2 = c(6, 6, 8, 8, 6, 4, 10, 16, 18, 18, 18, 16, 18, 18, 18),
                 x3 = c(4, 7, 7, 3, 8, 9, 9, 8, 7, 8, 9, 4, 9, 10, 13))

#fit logistic regression model
model <- glm(y~x1+x2+x3, data=df, family=binomial)

#view model summary
summary(model)

解决：

#进行相关性分析

#发现x1和x2几乎完全正相关；

#create correlation matrix
cor(df)

           y        x1        x2        x3
y  1.0000000 0.9675325 0.9675325 0.3610320
x1 0.9675325 1.0000000 1.0000000 0.3872889
x2 0.9675325 1.0000000 1.0000000 0.3872889
x3 0.3610320 0.3872889 0.3872889 1.0000000

#剔除强相关的变量之后再构建新的模型

#fit logistic regression model
model <- glm(y~x1+x3, data=df, family=binomial)

#view model summary
summary(model)

完整错误：

> #define data
> df <- data.frame(y = c(0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1),
+ x1 = c(3, 3, 4, 4, 3, 2, 5, 8, 9, 9, 9, 8, 9, 9, 9),
+ x2 = c(6, 6, 8, 8, 6, 4, 10, 16, 18, 18, 18, 16, 18, 18, 18),
+ x3 = c(4, 7, 7, 3, 8, 9, 9, 8, 7, 8, 9, 4, 9, 10, 13))
>
> #fit logistic regression model
> model <- glm(y~x1+x2+x3, data=df, family=binomial)
Warning message:
glm.fit:拟合機率算出来是数值零或一
>
> #view model summary
> summary(model)

Call:
glm(formula = y ~ x1 + x2 + x3, family = binomial, data = df)

Deviance Residuals:
Min 1Q Median 3Q Max
-1.372e-05 -2.110e-08 2.110e-08 2.110e-08 1.575e-05

Coefficients: (1 not defined because of singularities)
Estimate Std. Error z value Pr(>|z|)
(Intercept) -75.496 176487.031 0.000 1
x1 14.546 24314.459 0.001 1
x2 NA NA NA NA
x3 -2.258 20119.863 0.000 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 2.0728e+01 on 14 degrees of freedom
Residual deviance: 5.1523e-10 on 12 degrees of freedom
AIC: 6

Number of Fisher Scoring iterations: 24