| Title: | Optimal Binning of Continuous and Categorical Variables |
|---|---|
| Description: | Tool for easy and efficient discretization of continuous and categorical data. The package calculates the most optimal binning of a given explanatory variable with respect to a user-specified target variable. The purpose is to assign a unique Weight-of-Evidence value to each of the calculated binpoints in order to recode the original variable. The package allows users to impose certain restrictions on the functional form on the resulting binning while maximizing the overall information value in the original data. The package is well suited for logistic scoring models where input variables may be subject to restrictions such as linearity by e.g. regulatory authorities. An excellent source describing in detail the development of scorecards, and the role of Weight-of-Evidence coding in credit scoring is (Siddiqi 2006, ISBN: 978–0-471–75451–0). The package utilizes the discrete nature of decision trees and Isotonic Regression to accommodate the trade-off between flexible functional forms and maximum information value. |
| Authors: | Daniel Safai |
| Maintainer: | Daniel Safai <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 0.2.1 |
| Built: | 2025-02-11 02:40:15 UTC |
| Source: | https://github.com/cran/varbin |
Optimal binning of numerical variable
varbin(df, x, y, p=0.05, custom_vec=NA)varbin(df, x, y, p=0.05, custom_vec=NA)
df |
A data frame |
x |
String. Name of continuous variable in data frame. |
y |
String. Name of binary response variable (0,1) in data frame. |
p |
Percentage of records per bin. Default 5 pct. (0.05). This parameter only accepts values greater than 0.00 (0 pct.) and lower than 0.50 (50 pct.). |
custom_vec |
Numerical input vector with custom cutpoints. E.g. custom_vec=c(20, 50, 75) for a variable representing age, will result in the cutpoints [<20, <50, <75, >=75]. NA results in the default unrestricted (most optimal) binning. |
The command varbin generates a data frame with necessary info and utilities for binning. The user should save the output result so it can be used with e.g. varbin.plot, or varbin.convert.
# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform unrestricted binning result <- varbin(df, "age", "target") # Perform custom binning result2 <- varbin(df, "age", "target", custom_vec=c(30,40,60,75))# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform unrestricted binning result <- varbin(df, "age", "target") # Perform custom binning result2 <- varbin(df, "age", "target", custom_vec=c(30,40,60,75))
Generate new variable based on constructed binnings
varbin.convert(df, ivTable, x)varbin.convert(df, ivTable, x)
df |
A data frame |
ivTable |
Output from either varbin, varbin.factor, varbin.monotonic or varbin.kink. |
x |
String. Name of variable in data frame for which binninngs should be applied. |
The command varbin.convert appends a new variable named "WoE_[x]" to the data frame. The new variable consist of the Weight of Evidence values from the resulting binning.
# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Split train/test df_train <- df[1:5000, ] df_test <- df[5001:nrow(df), ] # Perform unrestricted binnings result <- varbin.factor(df_train, "educ", "target") result2 <- varbin(df_train, "age", "target") # Convert test data df_new <- varbin.convert(rbind(df_train, df_test), result,"educ") df_new <- varbin.convert(df_new, result2, "age")# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Split train/test df_train <- df[1:5000, ] df_test <- df[5001:nrow(df), ] # Perform unrestricted binnings result <- varbin.factor(df_train, "educ", "target") result2 <- varbin(df_train, "age", "target") # Convert test data df_new <- varbin.convert(rbind(df_train, df_test), result,"educ") df_new <- varbin.convert(df_new, result2, "age")
Binning of categorical variable
varbin.factor(df, x, y, custom_vec=NA)varbin.factor(df, x, y, custom_vec=NA)
df |
A data frame |
x |
String. Name of factor variable in data frame. |
y |
String. Name of binary response variable (0,1) in data frame. |
custom_vec |
Character input vector with custom cutpoints. E.g. custom_vec=c("STUDENT", "UNEMP,RETIRED", "EMPLOYED") for a variable representing occupation, will result in the cutpoints ["STUDENT", "UNEMP,RETIRED", "EMPLOYED"]. NA results in default binning (no binning) i.e. the cutpoints ["STUDENT", "UNEMP", "RETIRED", "EMPLOYED"] corresponding to the levels of the factor variable. |
The command varbin generates a data frame with necessary info and utilities for binning. The user should save the output result so it can be used with e.g. varbin.plot, or varbin.convert.
# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform unrestricted binning result <- varbin.factor(df, "educ", "target") # Perform custom binning result2 <- varbin.factor(df, "educ", "target", custom_vec=c("MSC,BSC,PHD", "SELF", "OTHER"))# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform unrestricted binning result <- varbin.factor(df, "educ", "target") # Perform custom binning result2 <- varbin.factor(df, "educ", "target", custom_vec=c("MSC,BSC,PHD", "SELF", "OTHER"))
Impose global/local extremum i.e. a kink restriction on binning of numerical variable (if possible)
varbin.kink(df, x, y, p=0.05)varbin.kink(df, x, y, p=0.05)
df |
A data frame |
x |
String. Name of continuous variable in data frame. |
y |
String. Name of binary response variable (0,1) in data frame. |
p |
Percentage of records per bin. Default 5 pct. (0.05). This parameter only accepts values greater than 0.00 (0 pct.) and lower than 0.50 (50 pct.). |
The command varbin.kink generates a data frame with necessary info and utilities for a variable where the binnings are restricted such that the functional form is characterized by having a global/local minimum/maximum i.e. a kink. The function will not work for variables where both a monotonically in- or decreasing functional form can't be imposed The user should save the output result so it can be used with e.g. varbin.plot, or varbin.convert.
# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform restricted binning - note the kink shape of the WoE values in the output result <- varbin.kink(df, "inc", "target")# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform restricted binning - note the kink shape of the WoE values in the output result <- varbin.kink(df, "inc", "target")
Monotonically in- or decreasing restriction on binning of numerical variable
varbin.monotonic(df, x, y, p=0.05, increase=F, decrease=F, auto=T)varbin.monotonic(df, x, y, p=0.05, increase=F, decrease=F, auto=T)
df |
A data frame |
x |
String. Name of continuous variable in data frame. |
y |
String. Name of binary response variable (0,1) in data frame. |
p |
Percentage of records per bin. Default 5 pct. (0.05). This parameter only accepts values greater than 0.00 (0 pct.) and lower than 0.50 (50 pct.). |
increase |
Logical (TRUE/FALSE). Whether to force an increasing monotonic functional form (if possible) |
decrease |
Logical (TRUE/FALSE). Whether to force a decreasing monotonic functional form (if possible) |
auto |
Logical (TRUE/FALSE). Whether to choose which of the two above is most optimal |
The command varbin generates a data frame with necessary info and utilities for a monotonically in- or decreasing functional form restriction imposed to the binning. The user should save the output result so it can be used with e.g. varbin.plot, or varbin.convert.
# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform monotonically restricted binning result <- varbin.monotonic(df, "inc", "target")# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform monotonically restricted binning result <- varbin.monotonic(df, "inc", "target")
Generate simple plot to visualize binning results
varbin.plot(ivTable)varbin.plot(ivTable)
ivTable |
Output from either varbin, varbin.factor, varbin.monotonic or varbin.kink. |
The command varbin.plot generates a simple plot with the Weight of Evidence values on the y-axis and the cutpoints/binnings on the x-axis. Gives a nice overview of the functional form and the relatioship between the explanatory variable and the dependent variable.
# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform restricted binning result <- varbin.kink(df, "inc", "target") # Plot result varbin.plot(result)# Set seed and generate data set.seed(1337) target <- as.numeric(runif(10000, 0, 1)<0.2) age <- round(rnorm(10000, 40, 15), 0) age[age<20] <- round(rnorm(sum(age<20), 40, 5), 0) age[age>95] <- round(rnorm(sum(age>95), 40, 5), 0) inc <- round(rnorm(10000, 100000, 10000), 0) educ <- sample(c("MSC", "BSC", "SELF", "PHD", "OTHER"), 10000, replace=TRUE) df <- data.frame(target=target, age=age, inc=inc, educ=educ) # Perform restricted binning result <- varbin.kink(df, "inc", "target") # Plot result varbin.plot(result)