Clean Names of Dataset

The original names of the dataset are:

colnames(cc_defaults)

##  [1] "ID"                         "LIMIT_BAL"                 
##  [3] "SEX"                        "EDUCATION"                 
##  [5] "MARRIAGE"                   "AGE"                       
##  [7] "PAY_0"                      "PAY_2"                     
##  [9] "PAY_3"                      "PAY_4"                     
## [11] "PAY_5"                      "PAY_6"                     
## [13] "BILL_AMT1"                  "BILL_AMT2"                 
## [15] "BILL_AMT3"                  "BILL_AMT4"                 
## [17] "BILL_AMT5"                  "BILL_AMT6"                 
## [19] "PAY_AMT1"                   "PAY_AMT2"                  
## [21] "PAY_AMT3"                   "PAY_AMT4"                  
## [23] "PAY_AMT5"                   "PAY_AMT6"                  
## [25] "default payment next month"

Let’s clean the names of the dataset doing:

cc_defaults <- clean_names(cc_defaults)

The dataset names are now:

colnames(cc_defaults)

##  [1] "id"                         "limit_bal"                 
##  [3] "sex"                        "education"                 
##  [5] "marriage"                   "age"                       
##  [7] "pay_0"                      "pay_2"                     
##  [9] "pay_3"                      "pay_4"                     
## [11] "pay_5"                      "pay_6"                     
## [13] "bill_amt1"                  "bill_amt2"                 
## [15] "bill_amt3"                  "bill_amt4"                 
## [17] "bill_amt5"                  "bill_amt6"                 
## [19] "pay_amt1"                   "pay_amt2"                  
## [21] "pay_amt3"                   "pay_amt4"                  
## [23] "pay_amt5"                   "pay_amt6"                  
## [25] "default_payment_next_month"

Features Transformation

In the description of the dataset we see that some variables like sex, education and marriage are categorical, although they are coded as numerical. To see why is that misleading, let’s consider sex, which is equal to 1 for males and 2 for females. This labelling is arbitrary, so instead of describing a property larger for women than for men, is splitting the dataset into two different categories. We need to turn these variables into factors, and rename the levels of these factors adequately. This can be done with forcats::fct_recode.

cc_defaults <- cc_defaults |>
  mutate(sex = factor(sex)) |>
  mutate(sex = fct_recode(sex, male = "1", female = "2"))

The transformation of education is more complex. From the summary, we observe that goes from 0 to 6, but it is defined from 1 to 4. Let’s count how many observations have for each level using a janitor::tabyl function:

cc_defaults |>
  tabyl(education) |>
  adorn_pct_formatting() |>
  kbl() |>
  kable_styling(full_width = FALSE)

The proportion of odd values of the variable is small, so it is safer to group them into an unknown level:

cc_defaults <- cc_defaults |>
  mutate(education = factor(education)) |>
  mutate(education = fct_recode(education, graduate = "1", university = "2", high_school = "3", others = "4", unknown = "0", unknown = "5", unknown = "6"))

The resulting levels of the factor are:

levels(cc_defaults$education)

## [1] "unknown"     "graduate"    "university"  "high_school" "others"

And the new counting of levels is:

cc_defaults |>
  tabyl(education) |>
  adorn_pct_formatting() |>
  kbl() |>
  kable_styling(full_width = FALSE)

Something similar happens with the ’marriage` variable, which has more levels than the ones defined in Yeh & Lien (2009):

cc_defaults |>
  tabyl(marriage) |>
  adorn_pct_formatting() |>
  kbl() |>
  kable_styling(full_width = FALSE)

When transforming marriage into a factor, I will be coding the zero level as unknown:

cc_defaults <- cc_defaults |>
  mutate(marriage = factor(marriage)) |>
  mutate(marriage = fct_recode(marriage, unknown = "0", unknown = "3", married = "1", single = "2"))

Transforming the Target Variable

The target variable default_payment_next_month must be transformed into a factor, with the positive case as first level. I will use dplyr::rename to give the variable a shorter name and fct_relevel to reorder the levels of variables.

cc_defaults <- cc_defaults |>
  rename("default" = default_payment_next_month) |>
  mutate(default = factor(default)) |>
  mutate(default = fct_relevel(default, "1")) |>
  mutate(default = fct_recode(default, yes = "1", "no" = "0"))

Counting the observations of each level, we observe that the dataset is imbalanced, as there are much more observations with no default:

cc_defaults |>
  tabyl(default) |>
  adorn_pct_formatting() |>
  kbl() |>
  kable_styling(full_width = FALSE)

Rename Features

The family of pay_*, bill_amt* and pay_amt* variables are labelled inconsistently, and the names do not provide information about the month they are describing. We are trying to predict if a customer will default in October by examining her/his behavior in the previous six months, April to September. So I will relabel the variables accordingly using dplyr::rename:

cc_defaults <- cc_defaults |>
  rename("pay_sep" = pay_0,
         "pay_aug" = pay_2,
         "pay_jul" = pay_3,
         "pay_jun" = pay_4,
         "pay_may" = pay_5,
         "pay_apr" = pay_6,
         "bill_amt_sep" = bill_amt1,
         "bill_amt_aug" = bill_amt2,
         "bill_amt_jul" = bill_amt3,
         "bill_amt_jun" = bill_amt4,
         "bill_amt_may" = bill_amt5,
         "bill_amt_apr" = bill_amt6,
         "pay_amt_sep" = pay_amt1,
         "pay_amt_aug" = pay_amt2,
         "pay_amt_jul" = pay_amt3,
         "pay_amt_jun" = pay_amt4,
         "pay_amt_may" = pay_amt5,
         "pay_amt_apr" = pay_amt6)

Significance of pay Variables

The original meaning of the pay_apr:pay_sep variables is:

X6 - X11: History of past payment. We tracked the past monthly payment records (from April to September, 2005) as follows: X6 = the repayment status in September, 2005; X7 = the repayment status in August, 2005; . . .;X11 = the repayment status in April, 2005. The measurement scale for the repayment status is: -1 = pay duly; 1 = payment delay for one month; 2 = payment delay for two months; . . .; 8 = payment delay for eight months; 9 = payment delay for nine months and above.

Nevertheless, we observe that there are other variable values around. Let’s see for instance pay_sep:

cc_defaults |>
  tabyl(pay_sep) |>
  adorn_pct_formatting() |>
  kbl() |>
  kable_styling(full_width = FALSE)

This fact has caught the attention of other users. In a Kaggle discussion forum, we are informed that the authors of the paper provided further clarification about the meaning of this variable:

-2: No consumption; -1: Paid in full; 0: The use of revolving credit; 1 = payment delay for one month; 2 = payment delay for two months; . . .; 8 = payment delay for eight months; 9 = payment delay for nine months and above.

This means that the pay_* variables have a different meaning when they are positive than when they are zero or negative. Further analysis must take this into account.

The Importance of Data Cleaning

Data cleaning is a relevant tasks in data analysis. As quoted in the janitor website:

Data scientists, according to interviews and expert estimates, spend from 50 percent to 80 percent of their time mired in this more mundane labor of collecting and preparing unruly digital data, before it can be explored for useful nuggets.

The skimr::skim function helps to take a first look of a dataset with an exhaustive summary report. The janitor package helps to simplify this cumbersome task in the context of R, reproducing with janitor::tabyl some of the results of counting the number of elements of each level using Excel dynamic tables. The use of a programming language like R helps to make this data cleaning task reproducible.

References

• default of credit card clients Data Set https://archive.ics.uci.edu/ml/datasets/default+of+credit+card+clients# UCI Machine Learning Repository.
• Default of Credit Card Clients Dataset: Answer to missing values (-2, 0) not in measurement scale. https://www.kaggle.com/datasets/uciml/default-of-credit-card-clients-dataset/discussion/34608
• janitor website https://sfirke.github.io/janitor/index.html
• Yeh, I. C., & Lien, C. H. (2009). The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients. Expert Systems with Applications, 36(2), 2473-2480. https://doi.org/10.1016/j.eswa.2007.12.020

Session Info

## R version 4.3.0 (2023-04-21)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Linux Mint 21.1
## 
## Matrix products: default
## BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.10.0 
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
## 
## locale:
##  [1] LC_CTYPE=es_ES.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=es_ES.UTF-8        LC_COLLATE=es_ES.UTF-8    
##  [5] LC_MONETARY=es_ES.UTF-8    LC_MESSAGES=es_ES.UTF-8   
##  [7] LC_PAPER=es_ES.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=es_ES.UTF-8 LC_IDENTIFICATION=C       
## 
## time zone: Europe/Madrid
## tzcode source: system (glibc)
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] kableExtra_1.3.4 skimr_2.1.5      janitor_2.2.0    readxl_1.4.2    
##  [5] lubridate_1.9.2  forcats_1.0.0    stringr_1.5.0    dplyr_1.1.2     
##  [9] purrr_1.0.1      readr_2.1.4      tidyr_1.3.0      tibble_3.2.1    
## [13] ggplot2_3.4.2    tidyverse_2.0.0 
## 
## loaded via a namespace (and not attached):
##  [1] sass_0.4.5        utf8_1.2.3        generics_0.1.3    xml2_1.3.3       
##  [5] blogdown_1.16     stringi_1.7.12    hms_1.1.3         digest_0.6.31    
##  [9] magrittr_2.0.3    evaluate_0.20     grid_4.3.0        timechange_0.2.0 
## [13] bookdown_0.33     fastmap_1.1.1     cellranger_1.1.0  jsonlite_1.8.4   
## [17] httr_1.4.5        rvest_1.0.3       fansi_1.0.4       viridisLite_0.4.1
## [21] scales_1.2.1      jquerylib_0.1.4   cli_3.6.1         rlang_1.1.0      
## [25] munsell_0.5.0     base64enc_0.1-3   repr_1.1.6        withr_2.5.0      
## [29] cachem_1.0.7      yaml_2.3.7        tools_4.3.0       tzdb_0.3.0       
## [33] colorspace_2.1-0  webshot_0.5.4     vctrs_0.6.2       R6_2.5.1         
## [37] lifecycle_1.0.3   snakecase_0.11.0  pkgconfig_2.0.3   pillar_1.9.0     
## [41] bslib_0.4.2       gtable_0.3.3      glue_1.6.2        systemfonts_1.0.4
## [45] highr_0.10        xfun_0.39         tidyselect_1.2.0  rstudioapi_0.14  
## [49] knitr_1.42        htmltools_0.5.5   svglite_2.1.1     rmarkdown_2.21   
## [53] compiler_4.3.0