Introducing the custom binning App

Jurriaan Nagelkerke
23-8-2015

Value of binning for predictive modeling

Binning is very useful in predictive modeling

To identify how variables are related
To visualize correlations
To get rid of outliers and missing values

Below, an example is included that shows how easy it is to spot a correlation between sepal length and the specy “Satosa” in the Iris dataset in the binned version of sepal with.

library(datasets)
data(iris)

iris_ext<-iris
iris_ext$ind_setosa<-iris$Species=="setosa"

quantiles<-quantile(iris_ext$Sepal.Length,probs = (1:5)/5,na.rm = TRUE)

iris_ext$Sepal.Length_binned <- 5
iris_ext$Sepal.Length_binned[iris_ext$Sepal.Length<=quantiles[4]]<-4
iris_ext$Sepal.Length_binned[iris_ext$Sepal.Length<=quantiles[3]]<-3
iris_ext$Sepal.Length_binned[iris_ext$Sepal.Length<=quantiles[2]]<-2
iris_ext$Sepal.Length_binned[iris_ext$Sepal.Length<=quantiles[1]]<-1

# Stacked Bar Plot with Colors and Legend
counts <- table(iris_ext$ind_setosa, iris_ext$Sepal.Length_binned)
barplot(counts, main="Species Setosa (Y/N) per Sepal length bin",
        xlab="Sepal length (binned)", col=c("red","green"),
        legend = rownames(counts))

plot of chunk unnamed-chunk-1

Features of the app

The following features make the app very usefull:

You can select a data source (in the current version: MLBattend - an R example set from the UsingR package, see http://www.inside-r.org/packages/cran/UsingR/docs/MLBattend for details)
You can select all numeric fields within the dataset
You can select the number of equal-sized bins you would like to have
You can evaluatie the cutoff points visually since the points for the bins are visualized in the histogram graph of the original variable
You can adjust the exact cutoff values to have more meaningfull cutoffs. For example: 1.000.000 as a cutoff instead of 995289.29
You can copy and paste the resulting code to create the binned variable. In case you've adjusted the cutoffs, these adjusted values are used; otherwise the calculated cutoffs are used.

How to get to the app

The app can be found on the following URL: https://jurrr.shinyapps.io/course-project

If you would like to know more on the source code (server.R and ui.R): https://github.com/jurrr/DevelopingDataProducts_CourseProject

Happy binning!!

Thanks for trying out this app! Hope it suits your goals in transforming your analysis data!

beybey