如何从R中的数据框中删除异常值？

Question

I have a data frame with 25 Variables. I want to remove the outliers from it.

I have searched SO forum and found that there are custom kind of solutions people are proposing for different posts.

Is there some standard R function that removes the outliers from the data?

Here are two functions I found from search. How good they are OR is there some standard same kind of better solution to achieve this in R in any package.

OR a function which I pass one column as argument & it returns outliers removed data.

remove_outliers: Link 1

Removing outliers - quick & dirty: Link 2

EDIT

The data in my data frame contains continuous data from two sources ie weather and ground. From weather, the predictors are temperature, humidity, wind, rain, solar radiation. And from ground are groundwater and soil moisture. I want to find a relation between soil moisture and other variables. I am analysing data using different models. Now I want to se the results after removing the outliers from data.

EDIT I used and edited code from one of the tutorials I added reference above. It's working fine when there are some outliers in the data. But it raises error when there are no. How to correct this.

Here is code:

outlier_rem<-Data_combined #data-frame with 25 var, few have outliers

#removong outliers from the column

outliers <- boxplot(outlier_rem$var1, plot=FALSE)$out
#print(outliers)
#ol<-outlier_rem[which(outlier_rem$var1 %in% outliers),]
ol<-outlier_rem[-which(outlier_rem$var1 %in% outliers),]

dim(ol)
boxplot(ol)

Here is error msg when ol returns 0 vale.

> dim(ol)
[1]  0 25
> boxplot(ol)
no non-missing arguments to min; returning Infno non-missing arguments to max; returning -InfError in plot.window(xlim = xlim, ylim = ylim, log = log, yaxs = pars$yaxs) : 
  need finite 'ylim' values

Answer 1

I use the Chebyshev's inequality as a criterion for dropping extreme values. It has the advantage that it holds true in many probablility distributions. The rule states tha no more than 1/k^2 of the values can be more than k standard deviations away from the mean. For example:

> x <- rchisq(1000, 13)
> 
> mean(x)
[1] 12.83906
> sd(x)
[1] 4.93234
> 
> Ndesv <- 5
> 
> x[x > (mean(x) + Ndesv * sd(x))]
[1] 38.7575
> 
> Conf <- (1 - 1 / Ndesv^2)
> print(Conf)
[1] 0.96
> 

Hope it helps you.