README.md

In jbkunst/irks: Tools for credit risk modelling

irks

Is a set of tools for credit risk modelling.

Installation

You can install the released version of irks from CRAN with:

install.packages("irks")

And the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("jbkunst/irks")

Descriptive

library(irks)
library(dplyr)
data("german_credit")

german_credit <- select(german_credit, good_bad, status_of_existing_checking_account,
                        duration_in_month, savings_account_or_bonds, credit_amount,
                        purpose, age_in_years)

duni <- describe(german_credit)
duni
#> $numeric
#> # A tibble: 4 x 10
#>   variable n_unique na_percent   min   mean median   max Variance skewness
#>   <chr>       <dbl>      <dbl> <dbl>  <dbl>  <dbl> <dbl>    <dbl>    <dbl>
#> 1 good_bad        2          0     0    0.7     1      1  2.10e-1   -0.873
#> 2 duratio~       33          0     4   20.9    18     72  1.45e+2    1.09 
#> 3 credit_~      921          0   250 3271.   2320. 18424  7.97e+6    1.95 
#> 4 age_in_~       53          0    19   35.5    33     75  1.29e+2    1.02 
#> # ... with 1 more variable: kurtosis <dbl>
#> 
#> $non_numeric
#> # A tibble: 3 x 4
#>   variable                            n_unique na_percent   hhi
#>   <chr>                                  <dbl>      <dbl> <dbl>
#> 1 status_of_existing_checking_account        4          0 0.307
#> 2 savings_account_or_bonds                   5          0 0.414
#> 3 purpose                                   10          0 0.189

dbiv <- describe_bivariate(german_credit, target = good_bad, verbose = FALSE)

arrange(dbiv, desc(iv))
#> # A tibble: 6 x 11
#>   variable                   iv   hhi n_unique class   monotonous na_level
#>   <chr>                   <dbl> <dbl>    <int> <chr>   <lgl>      <chr>   
#> 1 status_of_existing_ch~ 0.635  0.356        3 charac~ NA         (Missin~
#> 2 duration_in_month      0.229  0.484        3 integer TRUE       (Missin~
#> 3 savings_account_or_bo~ 0.189  0.585        2 charac~ NA         (Missin~
#> 4 credit_amount          0.175  0.335        4 integer FALSE      (Missin~
#> 5 purpose                0.142  0.523        2 charac~ NA         (Missin~
#> 6 age_in_years           0.0732 0.692        2 integer TRUE       (Missin~
#> # ... with 4 more variables: separate_missing <lgl>, dict <list>,
#> #   summary <list>, binning <list>

Binning

x <- german_credit$duration_in_month
y <- german_credit$good_bad

bin <- binning(x, y)
bin
#> binning object:
#> 
#>  Information Value: 0.229   medium 
#>  Herfindahl-Hirschman Index: 0.484   high concentration 
#>  Number of categories: 3 
#> 
#>  Summary table:
#> # A tibble: 3 x 13
#>   x             n percent target_n target_rate target_percent non_target_n
#>   <fct>     <int>   <dbl>    <dbl>       <dbl>          <dbl>        <dbl>
#> 1 [-Inf,11]   180    0.18      153       0.85           0.219           27
#> 2 (11,33]     650    0.65      459       0.706          0.656          191
#> 3 (33, Inf]   170    0.17       88       0.518          0.126           82
#> # ... with 6 more variables: non_target_percent <dbl>, odds <dbl>,
#> #   woe <dbl>, iv <dbl>, lower <dbl>, upper <dbl>
#> 
#>  Object values: dict, summary, separate_missing, iv, hhi, monotonous, na_level, class

# plot(bin)

Model helpers

var_importance

german_credit <- mutate_if(german_credit, is.character, as.factor)

rf <- randomForest::randomForest(factor(good_bad) ~ ., data = german_credit)
var_importance(rf)
#> # A tibble: 6 x 2
#>   variable                            importance
#>   <chr>                                    <dbl>
#> 1 credit_amount                            108. 
#> 2 age_in_years                              79.1
#> 3 duration_in_month                         66.7
#> 4 status_of_existing_checking_account       58.4
#> 5 purpose                                   52.3
#> 6 savings_account_or_bonds                  31.4

rrf <- RRF::RRF(factor(good_bad) ~ ., data = german_credit)
var_importance(rrf)
#> # A tibble: 6 x 2
#>   variable                            importance
#>   <chr>                                    <dbl>
#> 1 credit_amount                            127. 
#> 2 age_in_years                              86.5
#> 3 duration_in_month                         66.3
#> 4 status_of_existing_checking_account       60.8
#> 5 purpose                                   49.5
#> 6 savings_account_or_bonds                  29.1

library(partykit)
ct <- ctree(Species ~ ., data = iris)
ct_rules(ct)
#>   node                                                          rule
#> 1    2                                           Petal.Length <= 1.9
#> 2    5 Petal.Length > 1.9 & Petal.Width <= 1.7 & Petal.Length <= 4.8
#> 3    6  Petal.Length > 1.9 & Petal.Width <= 1.7 & Petal.Length > 4.8
#> 4    7                        Petal.Length > 1.9 & Petal.Width > 1.7

jbkunst/irks documentation built on May 22, 2021, 2:09 p.m.