| Title: | Scoring Modeling and Optimal Binning |
|---|---|
| Description: | A set of functions to build a scoring model from beginning to end, leading the user to follow an efficient and organized development process, reducing significantly the time spent on data exploration, variable selection, feature engineering, binning and model selection among other recurrent tasks. The package also incorporates monotonic and customized binning, scaling capabilities that transforms logistic coefficients into points for a better business understanding and calculates and visualizes classic performance metrics of a classification model. |
| Authors: | Herman Jopia |
| Maintainer: | Herman Jopia <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 0.9 |
| Built: | 2025-02-27 03:00:26 UTC |
| Source: | https://github.com/cran/smbinning |
Optimal Binning categorizes a numeric characteristic into bins for ulterior usage in scoring modeling.
This process, also known as supervised discretization,
utilizes Recursive Partitioning to categorize
the numeric characteristic.
The especific algorithm is Conditional Inference Trees
which initially excludes missing values (NA) to compute the cutpoints, adding them back later in the
process for the calculation of the Information Value.
smbinning(df, y, x, p = 0.05)smbinning(df, y, x, p = 0.05)
df |
A data frame. |
y |
Binary response variable (0,1). Integer ( |
x |
Continuous characteristic. At least 5 different values. Value |
p |
Percentage of records per bin. Default 5% (0.05). This parameter only accepts values greater that 0.00 (0%) and lower than 0.50 (50%). |
The command smbinning generates and object containing the necessary info and utilities for binning.
The user should save the output result so it can be used
with smbinning.plot, smbinning.sql, and smbinning.gen.
# Load library and its dataset library(smbinning) # Load package and its data # Example: Optimal binning result=smbinning(df=smbsimdf1,y="fgood",x="cbs1") # Run and save result result$ivtable # Tabulation and Information Value result$iv # Information value result$bands # Bins or bands result$ctree # Decision tree# Load library and its dataset library(smbinning) # Load package and its data # Example: Optimal binning result=smbinning(df=smbsimdf1,y="fgood",x="cbs1") # Run and save result result$ivtable # Tabulation and Information Value result$iv # Information value result$bands # Bins or bands result$ctree # Decision tree
It gives the user the ability to create customized cutpoints.
smbinning.custom(df, y, x, cuts)smbinning.custom(df, y, x, cuts)
df |
A data frame. |
y |
Binary response variable (0,1). Integer ( |
x |
Continuous characteristic. At least 5 different values. Value |
cuts |
Vector with the cutpoints selected by the user. It does not have a default so user must define it. |
The command smbinning.custom generates and object containing the necessary info and utilities for binning.
The user should save the output result so it can be used
with smbinning.plot, smbinning.sql, and smbinning.gen.
# Load library and its dataset library(smbinning) # Load package and its data # Custom cutpoints using percentiles (20% each) cbs1cuts=as.vector(quantile(smbsimdf1$cbs1, probs=seq(0,1,0.2), na.rm=TRUE)) # Quantiles cbs1cuts=cbs1cuts[2:(length(cbs1cuts)-1)] # Remove first (min) and last (max) values # Example: Customized binning result=smbinning.custom(df=smbsimdf1,y="fgood",x="cbs1",cuts=cbs1cuts) # Run and save result$ivtable # Tabulation and Information Value# Load library and its dataset library(smbinning) # Load package and its data # Custom cutpoints using percentiles (20% each) cbs1cuts=as.vector(quantile(smbsimdf1$cbs1, probs=seq(0,1,0.2), na.rm=TRUE)) # Quantiles cbs1cuts=cbs1cuts[2:(length(cbs1cuts)-1)] # Remove first (min) and last (max) values # Example: Customized binning result=smbinning.custom(df=smbsimdf1,y="fgood",x="cbs1",cuts=cbs1cuts) # Run and save result$ivtable # Tabulation and Information Value
It shows basic statistics for each characteristic in a data frame. The report includes:
Field: Field name.
Type: Factor, numeric, integer, other.
Recs: Number of records.
Miss: Number of missing records.
Min: Minimum value.
Q25: First quartile. It splits off the lowest 25% of data from the highest 75%.
Q50: Median or second quartile. It cuts data set in half.
Avg: Average value.
Q75: Third quartile. It splits off the lowest 75% of data from the highest 25%.
Max: Maximum value.
StDv: Standard deviation of a sample.
Neg: Number of negative values.
Pos: Number of positive values.
OutLo: Number of outliers. Records below Q25-1.5*IQR, where IQR=Q75-Q25.
OutHi: Number of outliers. Records above Q75+1.5*IQR, where IQR=Q75-Q25.
smbinning.eda(df, rounding = 3, pbar = 1)smbinning.eda(df, rounding = 3, pbar = 1)
df |
A data frame. |
rounding |
Optional parameter to define the decimal points shown in the output table. Default is 3. |
pbar |
Optional parameter that turns on or off a progress bar. Default value is 1. |
The command smbinning.eda generates two data frames that list each characteristic
with basic statistics such as extreme values and quartiles;
and also percentages of missing values and outliers, among others.
# Load library and its dataset library(smbinning) # Load package and its data # Example: Exploratory data analysis of dataset smbinning.eda(smbsimdf1,rounding=3)$eda # Table with basic statistics smbinning.eda(smbsimdf1,rounding=3)$edapct # Table with basic percentages# Load library and its dataset library(smbinning) # Load package and its data # Example: Exploratory data analysis of dataset smbinning.eda(smbsimdf1,rounding=3)$eda # Table with basic statistics smbinning.eda(smbsimdf1,rounding=3)$edapct # Table with basic percentages
It generates a table with relevant metrics for all the categories of a given factor variable.
smbinning.factor(df, y, x, maxcat = 10)smbinning.factor(df, y, x, maxcat = 10)
df |
A data frame. |
y |
Binary response variable (0,1). Integer ( |
x |
A factor variable with at least 2 different values. Labesl with commas are not allowed. |
maxcat |
Specifies the maximum number of categories. Default value is 10.
Name of |
The command smbinning.factor generates and object containing the necessary info and utilities for binning.
The user should save the output result so it can be used
with smbinning.plot, smbinning.sql, and smbinning.gen.factor.
# Load library and its dataset library(smbinning) # Load package and its data # Binning a factor variable result=smbinning.factor(smbsimdf1,x="inc",y="fgood", maxcat=11) result$ivtable# Load library and its dataset library(smbinning) # Load package and its data # Binning a factor variable result=smbinning.factor(smbsimdf1,x="inc",y="fgood", maxcat=11) result$ivtable
It gives the user the ability to combine categories and create new attributes for a given characteristic.
Once these new attribues are created in a list (called groups), the funtion generates a table for
the uniques values of a given factor variable.
smbinning.factor.custom(df, y, x, groups)smbinning.factor.custom(df, y, x, groups)
df |
A data frame. |
y |
Binary response variable (0,1). Integer ( |
x |
A factor variable with at least 2 different values. Value |
groups |
Specifies customized groups created by the user.
Name of |
The command smbinning.factor.custom generates an object containing the necessary information
and utilities for binning.
The user should save the output result so it can be used
with smbinning.plot, smbinning.sql, and smbinning.gen.factor.
# Load library and its dataset library(smbinning) # Load package and its data # Example: Customized binning for a factor variable # Notation: Groups between double quotes result=smbinning.factor.custom( smbsimdf1,x="inc", y="fgood", c("'W01','W02'", # Group 1 "'W03','W04','W05'", # Group 2 "'W06','W07'", # Group 3 "'W08','W09','W10'")) # Group 4 result$ivtable# Load library and its dataset library(smbinning) # Load package and its data # Example: Customized binning for a factor variable # Notation: Groups between double quotes result=smbinning.factor.custom( smbsimdf1,x="inc", y="fgood", c("'W01','W02'", # Group 1 "'W03','W04','W05'", # Group 2 "'W06','W07'", # Group 3 "'W08','W09','W10'")) # Group 4 result$ivtable
It generates a data frame with a new predictive characteristic from a factor variable after applying
smbinning.factor or smbinning.factor.custom.
smbinning.factor.gen(df, ivout, chrname = "NewChar")smbinning.factor.gen(df, ivout, chrname = "NewChar")
df |
Dataset to be updated with the new characteristic. |
ivout |
An object generated after |
chrname |
Name of the new characteristic. |
A data frame with the binned version of the original characteristic.
# Load library and its dataset library(smbinning) # Load package and its data pop=smbsimdf1 # Set population train=subset(pop,rnd<=0.7) # Training sample # Binning a factor variable on training data result=smbinning.factor(train,x="home",y="fgood") # Example: Append new binned characteristic to population pop=smbinning.factor.gen(pop,result,"g1home") # Split training train=subset(pop,rnd<=0.7) # Training sample # Check new field counts table(train$g1home) table(pop$g1home)# Load library and its dataset library(smbinning) # Load package and its data pop=smbsimdf1 # Set population train=subset(pop,rnd<=0.7) # Training sample # Binning a factor variable on training data result=smbinning.factor(train,x="home",y="fgood") # Example: Append new binned characteristic to population pop=smbinning.factor.gen(pop,result,"g1home") # Split training train=subset(pop,rnd<=0.7) # Training sample # Check new field counts table(train$g1home) table(pop$g1home)
It generates a data frame with a new predictive characteristic after applying
smbinning or smbinning.custom.
smbinning.gen(df, ivout, chrname = "NewChar")smbinning.gen(df, ivout, chrname = "NewChar")
df |
Dataset to be updated with the new characteristic. |
ivout |
An object generated after |
chrname |
Name of the new characteristic. |
A data frame with the binned version of the original characteristic.
# Load library and its dataset library(smbinning) # Load package and its data pop=smbsimdf1 # Set population train=subset(pop,rnd<=0.7) # Training sample # Binning application for a numeric variable result=smbinning(df=train,y="fgood",x="dep") # Run and save result # Generate a dataset with binned characteristic pop=smbinning.gen(pop,result,"g1dep") # Check new field counts table(pop$g1dep)# Load library and its dataset library(smbinning) # Load package and its data pop=smbsimdf1 # Set population train=subset(pop,rnd<=0.7) # Training sample # Binning application for a numeric variable result=smbinning(df=train,y="fgood",x="dep") # Run and save result # Generate a dataset with binned characteristic pop=smbinning.gen(pop,result,"g1dep") # Check new field counts table(pop$g1dep)
It runs all the possible logistic models for a given set of characteristics (chr) and then rank them
from highest to lowest performance based on AIC.
Important Note: This function may take time depending on the datset size and number of variables used in it.
The user should run it at the end of the modeling process once variables have been pre-selected in previous steps.
smbinning.logitrank(y, chr, df)smbinning.logitrank(y, chr, df)
y |
Binary dependent variable. |
chr |
Vector with the characteristics (independent variables). |
df |
Data frame. |
The command smbinning.logitrank returns a table with the combination of characteristics
and their corresponding AIC and deviance. The table is ordered by AIC from lowest (best) to highest.
# Load library and its dataset library(smbinning) # Load package and its data # Example: Best combination of characteristics smbinning.logitrank(y="fgood",chr=c("chr1","chr2","chr3"),df=smbsimdf3)# Load library and its dataset library(smbinning) # Load package and its data # Example: Best combination of characteristics smbinning.logitrank(y="fgood",chr=c("chr1","chr2","chr3"),df=smbsimdf3)
It computes the classic performance metrics of a scoring model, including AUC, KS and all the relevant ones from the classification matrix at a specific threshold or cutoff.
smbinning.metrics(dataset, prediction, actualclass, cutoff = NA, report = 1, plot = "none", returndf = 0)smbinning.metrics(dataset, prediction, actualclass, cutoff = NA, report = 1, plot = "none", returndf = 0)
dataset |
Data frame. |
prediction |
Classifier. A value generated by a classification model (Must be numeric). |
actualclass |
Binary variable (0/1) that represents the actual class (Must be numeric). |
cutoff |
Point at wich the classifier splits (predicts) the actual class (Must be numeric). If not specified, it will be estimated by using the maximum value of Youden J (Sensitivity+Specificity-1). If not found in the data frame, it will take the closest lower value. |
report |
Indicator defined by user. 1: Show report (Default), 0: Do not show report. |
plot |
Specifies the plot to be shown for overall evaluation. It has three options: 'auc' shows the ROC curve, 'ks' shows the cumulative distribution of the actual class and its maximum difference (KS Statistic), and 'none' (Default). |
returndf |
Option for the user to save the data frame behind the metrics. 1: Show data frame, 0: Do not show (Default). |
The command smbinning.metrics returns a report with classic performance metrics of a classification model.
# Load library and its dataset library(smbinning) # Load package and its data # Example: Metrics Credit Score 1 smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=1) # Show report smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", cutoff=600, report=1) # User cutoff smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=0, plot="auc") # Plot AUC smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=0, plot="ks") # Plot KS # Save table with all details of metrics cbs1metrics=smbinning.metrics( dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=0, returndf=1) # Save metrics details# Load library and its dataset library(smbinning) # Load package and its data # Example: Metrics Credit Score 1 smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=1) # Show report smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", cutoff=600, report=1) # User cutoff smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=0, plot="auc") # Plot AUC smbinning.metrics(dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=0, plot="ks") # Plot KS # Save table with all details of metrics cbs1metrics=smbinning.metrics( dataset=smbsimdf1,prediction="cbs1",actualclass="fgood", report=0, returndf=1) # Save metrics details
It generates four plots after running and saving the output report from smbinning.metrics.
smbinning.metrics.plot(df, cutoff = NA, plot = "cmactual")smbinning.metrics.plot(df, cutoff = NA, plot = "cmactual")
df |
Data frame generated with |
cutoff |
Value of the classifier that splits the data between positive (>=) and negative (<). |
plot |
Plot to be drawn. Options are: 'cmactual' (default),'cmactualrates','cmmodel','cmmodelrates'. |
# Load library and its dataset library(smbinning) smbmetricsdf=smbinning.metrics(dataset=smbsimdf1, prediction="cbs1", actualclass="fgood", returndf=1) # Example 1: Plots based on optimal cutoff smbinning.metrics.plot(df=smbmetricsdf,plot='cmactual') # Example 2: Plots using user defined cutoff smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmactual') smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmactualrates') smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmmodel') smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmmodelrates')# Load library and its dataset library(smbinning) smbmetricsdf=smbinning.metrics(dataset=smbsimdf1, prediction="cbs1", actualclass="fgood", returndf=1) # Example 1: Plots based on optimal cutoff smbinning.metrics.plot(df=smbmetricsdf,plot='cmactual') # Example 2: Plots using user defined cutoff smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmactual') smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmactualrates') smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmmodel') smbinning.metrics.plot(df=smbmetricsdf,cutoff=600,plot='cmmodelrates')
It gives the user the ability to impose a monotonic trend for good/bad rates per bin.
smbinning.monotonic(df, y, x, p = 0.05)smbinning.monotonic(df, y, x, p = 0.05)
df |
A data frame. |
y |
Binary response variable (0,1). Integer ( |
x |
Continuous characteristic. At least 5 different values. Value |
p |
Percentage of records per bin. Default 5% (0.05). |
The command smbinning.monotonic generates and object containing the necessary info and utilities for binning.
The user should save the output result so it can be used
with smbinning.plot, smbinning.sql, and smbinning.gen.
# Load library and its dataset library(smbinning) # Load package and its data # Example 1: Monotonic Binning (Increasing Good Rate per Bin) smbinning(df=smbsimdf2,y="fgood2",x="chr2",p=0.05)$ivtable # Run regular binning smbinning.monotonic(df=smbsimdf2,y="fgood2",x="chr2",p=0.05)$ivtable # Run monotonic binning # Example 2: Monotonic Binning (Decreasing Good Rate per Bin) smbinning(df=smbsimdf2,y="fgood3",x="chr3",p=0.05)$ivtable # Run regular binning smbinning.monotonic(df=smbsimdf2,y="fgood3",x="chr3",p=0.05)$ivtable # Run monotonic binning# Load library and its dataset library(smbinning) # Load package and its data # Example 1: Monotonic Binning (Increasing Good Rate per Bin) smbinning(df=smbsimdf2,y="fgood2",x="chr2",p=0.05)$ivtable # Run regular binning smbinning.monotonic(df=smbsimdf2,y="fgood2",x="chr2",p=0.05)$ivtable # Run monotonic binning # Example 2: Monotonic Binning (Decreasing Good Rate per Bin) smbinning(df=smbsimdf2,y="fgood3",x="chr3",p=0.05)$ivtable # Run regular binning smbinning.monotonic(df=smbsimdf2,y="fgood3",x="chr3",p=0.05)$ivtable # Run monotonic binning
It generates plots for distribution, bad rate, and weight of evidence after running smbinning
and saving its output.
smbinning.plot(ivout, option = "dist", sub = "")smbinning.plot(ivout, option = "dist", sub = "")
ivout |
An object generated by binning. |
option |
Distribution ("dist"), Good Rate ("goodrate"), Bad Rate ("badrate"), and Weight of Evidence ("WoE"). |
sub |
Subtitle for the chart (optional). |
# Load library and its dataset library(smbinning) # Example 1: Numeric variable (1 page, 4 plots) result=smbinning(df=smbsimdf1,y="fgood",x="cbs1") # Run and save result par(mfrow=c(2,2)) boxplot(smbsimdf1$cbs1~smbsimdf1$fgood, horizontal=TRUE, frame=FALSE, col="lightgray",main="Distribution") mtext("Credit Score",3) smbinning.plot(result,option="dist",sub="Credit Score") smbinning.plot(result,option="badrate",sub="Credit Score") smbinning.plot(result,option="WoE",sub="Credit Score") par(mfrow=c(1,1)) # Example 2: Factor variable (1 plot per page) result=smbinning.factor(df=smbsimdf1,y="fgood",x="inc",maxcat=11) smbinning.plot(result,option="dist",sub="Income Level") smbinning.plot(result,option="badrate",sub="Income Level") smbinning.plot(result,option="WoE",sub="Income Level")# Load library and its dataset library(smbinning) # Example 1: Numeric variable (1 page, 4 plots) result=smbinning(df=smbsimdf1,y="fgood",x="cbs1") # Run and save result par(mfrow=c(2,2)) boxplot(smbsimdf1$cbs1~smbsimdf1$fgood, horizontal=TRUE, frame=FALSE, col="lightgray",main="Distribution") mtext("Credit Score",3) smbinning.plot(result,option="dist",sub="Credit Score") smbinning.plot(result,option="badrate",sub="Credit Score") smbinning.plot(result,option="WoE",sub="Credit Score") par(mfrow=c(1,1)) # Example 2: Factor variable (1 plot per page) result=smbinning.factor(df=smbsimdf1,y="fgood",x="inc",maxcat=11) smbinning.plot(result,option="dist",sub="Income Level") smbinning.plot(result,option="badrate",sub="Income Level") smbinning.plot(result,option="WoE",sub="Income Level")
Often models are developed using multiple periods in time for a number of reasons. For example, to avoid seasonality, to increase the size of the population, and some others. With a metrics like the Population Stability Index (PSI), users can check if there is a significant variation in the distribution of a certain feature by partition (usually time) using the first one as the reference.
smbinning.psi(df, y, x)smbinning.psi(df, y, x)
df |
Data frame. |
y |
Column name the indicates the different partitions. |
x |
Feature to be evaluated in terms of stability (It must be factor). |
Three crosstabs by feature and period that show the frequency (psicnt), percentage (psipct) and PSI (psimg), and a plot for the analyzed characteristic.
# Load library and its dataset library(smbinning) # Check stability for income smbinning.psi(df=smbsimdf1,y="period",x="inc")# Load library and its dataset library(smbinning) # Check stability for income smbinning.psi(df=smbsimdf1,y="period",x="inc")
It transforms the coefficients of a logistic regression into scaled points based on the following three parameters pre-selected by the analyst: PDO, Score, and Odds.
smbinning.scaling(logitraw, pdo = 20, score = 720, odds = 99)smbinning.scaling(logitraw, pdo = 20, score = 720, odds = 99)
logitraw |
Logistic regression (glm) that must have specified |
pdo |
Points to double the oods. |
score |
Score at which the desire |
odds |
Desired |
A scaled model from a logistic regression built with binned variables, the parameters used in the scaling process, the expected minimum and maximum score, and the original logistic model.
# Load library and its dataset library(smbinning) # Sampling pop=smbsimdf1 # Population train=subset(pop,rnd<=0.7) # Training sample # Generate binning object to generate variables smbcbs1=smbinning(train,x="cbs1",y="fgood") smbcbinq=smbinning.factor(train,x="cbinq",y="fgood") smbcblineut=smbinning.custom(train,x="cblineut",y="fgood",cuts=c(30,40,50)) smbpmt=smbinning.factor(train,x="pmt",y="fgood") smbtob=smbinning.custom(train,x="tob",y="fgood",cuts=c(1,2,3)) smbdpd=smbinning.factor(train,x="dpd",y="fgood") smbdep=smbinning.custom(train,x="dep",y="fgood",cuts=c(10000,12000,15000)) smbod=smbinning.factor(train,x="od",y="fgood") smbhome=smbinning.factor(train,x="home",y="fgood") smbinc=smbinning.factor.custom( train,x="inc",y="fgood", c("'W01','W02'","'W03','W04','W05'","'W06','W07'","'W08','W09','W10'")) pop=smbinning.gen(pop,smbcbs1,"g1cbs1") pop=smbinning.factor.gen(pop,smbcbinq,"g1cbinq") pop=smbinning.gen(pop,smbcblineut,"g1cblineut") pop=smbinning.factor.gen(pop,smbpmt,"g1pmt") pop=smbinning.gen(pop,smbtob,"g1tob") pop=smbinning.factor.gen(pop,smbdpd,"g1dpd") pop=smbinning.gen(pop,smbdep,"g1dep") pop=smbinning.factor.gen(pop,smbod,"g1od") pop=smbinning.factor.gen(pop,smbhome,"g1home") pop=smbinning.factor.gen(pop,smbinc,"g1inc") # Resample train=subset(pop,rnd<=0.7) # Training sample test=subset(pop,rnd>0.7) # Testing sample # Run logistic regression f=fgood~g1cbs1+g1cbinq+g1cblineut+g1pmt+g1tob+g1dpd+g1dep+g1od+g1home+g1inc modlogisticsmb=glm(f,data = train,family = binomial()) summary(modlogisticsmb) # Example: Scaling from logistic parameters to points smbscaled=smbinning.scaling(modlogisticsmb,pdo=20,score=720,odds=99) smbscaled$logitscaled # Scaled model smbscaled$minmaxscore # Expected minimum and maximum Score smbscaled$parameters # Parameters used for scaling summary(smbscaled$logitraw) # Extract of original logistic regression # Example: Generate score from scaled model pop1=smbinning.scoring.gen(smbscaled=smbscaled, dataset=pop) # Example Generate SQL code from scaled model smbinning.scoring.sql(smbscaled)# Load library and its dataset library(smbinning) # Sampling pop=smbsimdf1 # Population train=subset(pop,rnd<=0.7) # Training sample # Generate binning object to generate variables smbcbs1=smbinning(train,x="cbs1",y="fgood") smbcbinq=smbinning.factor(train,x="cbinq",y="fgood") smbcblineut=smbinning.custom(train,x="cblineut",y="fgood",cuts=c(30,40,50)) smbpmt=smbinning.factor(train,x="pmt",y="fgood") smbtob=smbinning.custom(train,x="tob",y="fgood",cuts=c(1,2,3)) smbdpd=smbinning.factor(train,x="dpd",y="fgood") smbdep=smbinning.custom(train,x="dep",y="fgood",cuts=c(10000,12000,15000)) smbod=smbinning.factor(train,x="od",y="fgood") smbhome=smbinning.factor(train,x="home",y="fgood") smbinc=smbinning.factor.custom( train,x="inc",y="fgood", c("'W01','W02'","'W03','W04','W05'","'W06','W07'","'W08','W09','W10'")) pop=smbinning.gen(pop,smbcbs1,"g1cbs1") pop=smbinning.factor.gen(pop,smbcbinq,"g1cbinq") pop=smbinning.gen(pop,smbcblineut,"g1cblineut") pop=smbinning.factor.gen(pop,smbpmt,"g1pmt") pop=smbinning.gen(pop,smbtob,"g1tob") pop=smbinning.factor.gen(pop,smbdpd,"g1dpd") pop=smbinning.gen(pop,smbdep,"g1dep") pop=smbinning.factor.gen(pop,smbod,"g1od") pop=smbinning.factor.gen(pop,smbhome,"g1home") pop=smbinning.factor.gen(pop,smbinc,"g1inc") # Resample train=subset(pop,rnd<=0.7) # Training sample test=subset(pop,rnd>0.7) # Testing sample # Run logistic regression f=fgood~g1cbs1+g1cbinq+g1cblineut+g1pmt+g1tob+g1dpd+g1dep+g1od+g1home+g1inc modlogisticsmb=glm(f,data = train,family = binomial()) summary(modlogisticsmb) # Example: Scaling from logistic parameters to points smbscaled=smbinning.scaling(modlogisticsmb,pdo=20,score=720,odds=99) smbscaled$logitscaled # Scaled model smbscaled$minmaxscore # Expected minimum and maximum Score smbscaled$parameters # Parameters used for scaling summary(smbscaled$logitraw) # Extract of original logistic regression # Example: Generate score from scaled model pop1=smbinning.scoring.gen(smbscaled=smbscaled, dataset=pop) # Example Generate SQL code from scaled model smbinning.scoring.sql(smbscaled)
After applying smbinning.scaling to the model, smbinning.scoring generates a data frame
with the final Score and additional fields with the points assigned to each characteristic so the user
can see how the final score is calculated. Example shown on smbinning.scaling section.
smbinning.scoring.gen(smbscaled, dataset)smbinning.scoring.gen(smbscaled, dataset)
smbscaled |
Object generated using |
dataset |
A data frame. |
The command smbinning.scoring generates a data frame with the final scaled Score and its
corresponding scaled weights per characteristic.
After applying smbinning.scaling to the model, smbinning.scoring.sql generates a SQL code
that creates and updates all variables present in the scaled model. Example shown on smbinning.scaling section.
smbinning.scoring.sql(smbscaled)smbinning.scoring.sql(smbscaled)
smbscaled |
Object generated using |
The command smbinning.scoring.sql generates a SQL code to implement the model the model in SQL.
It outputs a SQL code to facilitate the generation of new binned characetristic in a SQL environment. User must define table and new characteristic name.
smbinning.sql(ivout)smbinning.sql(ivout)
ivout |
An object generated by |
A text with the SQL code for binning.
# Load library and its dataset library(smbinning) # Example 1: Binning a numeric variable result=smbinning(df=smbsimdf1,y="fgood",x="cbs1") # Run and save result smbinning.sql(result) # Example 2: Binning for a factor variable result=smbinning.factor(df=smbsimdf1,x="inc",y="fgood",maxcat=11) smbinning.sql(result) # Example 3: Customized binning for a factor variable result=smbinning.factor.custom( df=smbsimdf1,x="inc",y="fgood", c("'W01','W02'","'W03','W04','W05'", "'W06','W07'","'W08','W09','W10'")) smbinning.sql(result)# Load library and its dataset library(smbinning) # Example 1: Binning a numeric variable result=smbinning(df=smbsimdf1,y="fgood",x="cbs1") # Run and save result smbinning.sql(result) # Example 2: Binning for a factor variable result=smbinning.factor(df=smbsimdf1,x="inc",y="fgood",maxcat=11) smbinning.sql(result) # Example 3: Customized binning for a factor variable result=smbinning.factor.custom( df=smbsimdf1,x="inc",y="fgood", c("'W01','W02'","'W03','W04','W05'", "'W06','W07'","'W08','W09','W10'")) smbinning.sql(result)
It gives the user the ability to calculate, in one step, the IV for each characteristic of the dataset. This function also shows a progress bar so the user can see the status of the process.
smbinning.sumiv(df, y)smbinning.sumiv(df, y)
df |
A data frame. |
y |
Binary response variable (0,1). Integer ( |
The command smbinning.sumiv generates a table that lists each characteristic
with its corresponding IV for those where the calculation is possible, otherwise it will generate a
missing value (NA).
# Load library and its dataset library(smbinning) # Test sample test=subset(smbsimdf1,rnd>0.9) # Training sample test$rnd=NULL # Example: Information Value Summary testiv=smbinning.sumiv(test,y="fgood") testiv # Example: Plot of Information Value Summary smbinning.sumiv.plot(testiv)# Load library and its dataset library(smbinning) # Test sample test=subset(smbsimdf1,rnd>0.9) # Training sample test$rnd=NULL # Example: Information Value Summary testiv=smbinning.sumiv(test,y="fgood") testiv # Example: Plot of Information Value Summary smbinning.sumiv.plot(testiv)
It gives the user the ability to plot the Information Value by characteristic.
The chart only shows characteristics with a valid IV.
Example shown on smbinning.sumiv section.
smbinning.sumiv.plot(sumivt, cex = 0.9)smbinning.sumiv.plot(sumivt, cex = 0.9)
sumivt |
A data frame saved after |
cex |
Optional parameter for the user to control the font size of the characteristics displayed on the chart. The default value is 0.9 |
The command smbinning.sumiv.plot returns a plot that shows the IV
for each numeric and factor characteristic in the dataset.
A simulated dataset where the target variable is fgood, which represents the binary status of default (0) and not default (1).
Data frame with 2,500 rows and 22 columns with 500 defaults.
fgood: Default (0), Not Default (1).
cbs1: Credit quality index (1-100).
cbs2: Profitability index (1-100).
cbinq: Number of inquiries.
cbline: Number of credit lines.
cbterm: Number of term loans.
cblineut: Line utilization (0-100).
cbtob: Number of years on file.
cbdpd: Indicator of days past due on bureau (Yes, No).
cbnew: Number of new loans.
pmt: Type of payment (M: Manual, A: Autopay, P: Payroll).
tob: Time on books (Years).
dpd: Level of delinquency (No, Low, High).
dep: Amount of deposits.
dc: Number of transactions.
od: Number of overdrafts.
home: Home ownership indicator (Yes, No).
inc: Level of income.
dd: Number of electronic transfers.
online: Indicator of online activity (Yes, No).
rnd: Random number to select testing and training samples.
period: Factor that indicates the year/month of the data (Based on rnd).
A simulated dataset used to illustrate the application of monotonic binning.
Data frame with 2,500 rows and 6 columns.
fgood1: Default (0), Not Default (1) for Numeric Variable 1.
chr1: Numeric variable 1.
fgood2: Default (0), Not Default (1) for Numeric Variable 2.
chr2: Numeric variable 2.
fgood3: Default (0), Not Default (1) for Numeric Variable 3.
chr3: Numeric variable 3.
A simulated dataset used to illustrate the application of model ranking.
Data frame with 1,000 rows and 4 columns.
fgood1: Default (0), Not Default (1) for Numeric Variable 1.
chr1: Numeric variable 1.
chr2: Numeric variable 2.
chr3: Numeric variable 3.