Data Set Information

The German credit scoring data is a dataset provided by Prof. Hogmann in the file german.data. The data set has information about 1000 individuals, on the basis of which they have been classified as risky or not. The variable response in the dataset corresponds to the risk label, 1 has been classified as bad and 2 has been classified as good. In our analysis, we have changed these labels to 1 and 0, 0 corresponding to a good credit record and 1 corresponding to a bad one (positive class).

Libraries Used

library(knitr)
library(dplyr)
library(tidyr)
library(reshape2)
library(RColorBrewer)
library(GGally)
library(ggplot2)

Importing Data and Column Names

german_credit = read.table("http://archive.ics.uci.edu/ml/machine-learning-databases/statlog/german/german.data")

colnames(german_credit) = c("chk_acct", "duration", "credit_his", "purpose", 
                            "amount", "saving_acct", "present_emp", "installment_rate", "sex", "other_debtor", 
                            "present_resid", "property", "age", "other_install", "housing", "n_credits", 
                            "job", "n_people", "telephone", "foreign", "response")

german_credit$response = german_credit$response - 1
german_credit$response <- as.factor(german_credit$response)

Attributes There is a total on 21 attributes in the dataset. Their descriptions and details have been tabulated below: 1. Status of existing checking account. 2. Duration in month 3. Credit history 4. Purpose 5. Credit amount 6. Savings account/bonds 7: Present employment since 8. Installment rate in percentage of disposable income 9. Personal status and sex 10. Other debtors / guarantors 11. Present residence since 12. Property 13. Age in years 14. Other installment plans 15. Housing 16. Number of existing credits at this bank 17. Job 18. Number of people being liable to provide maintenance for 19. Telephone 20. foreign worker

Summary Statistics

kable(summary(german_credit))

We take the summary statistics of the dataset, the dataset has a total of 1000 observations with 21 variables, out of which 8 are numerical variables including the response and 13 are categorical variables with various levels. The summary statistics for the variables have been presented

Exploratory Data Analysis of Continuous Data

amount.mean = german_credit %>% select(amount, response) %>%group_by(response) %>% summarise( m =mean(amount))
duration.mean = german_credit %>% select(duration, response) %>%group_by(response) %>% summarise( m =mean(duration))
duration.mean = german_credit %>% select(duration, response) %>%group_by(response) %>% summarise(m =mean(duration))

ggplot(german_credit, aes(duration, fill=response)) + 
  geom_density(alpha=.5) + geom_vline(data=german_credit, aes(xintercept=duration.mean[,2],  colour=response),
                                      linetype="dashed", size=1)

ggplot(german_credit,aes(x=installment_rate, fill=response)) +
  geom_histogram()

ggplot(german_credit, aes(x=installment_rate, fill=response)) + 
  geom_histogram(position = "dodge") + 
  scale_y_log10() + theme(axis.text.x=element_text(angle=45, hjust=1)) + ylab("Log Value") + xlab("Variables")

test.m = german_credit[,c(2,5,8,13,16,18,21)]
test.m$response <- as.numeric(test.m$response)
ggplot(melt(german_credit[,c(2,21)]), aes(x = variable, y = value, fill = response)) + geom_boxplot() + xlab("duration")

ggplot(melt(german_credit[,c(5,21)]), aes(x = variable, y = value, fill = response)) + 
geom_boxplot() + xlab("amount")

ggplot(melt(german_credit[,c(8,21)]), aes(x = variable, y = value, fill = response)) + 
geom_boxplot() + xlab("installment_rate")

ggplot(melt(german_credit[,c(13,21)]), aes(x = variable, y = value, fill = response)) + 
geom_boxplot() + xlab("age")

ggplot(melt(german_credit[,c(16,21)]), aes(x = variable, y = value, fill = response)) + 
geom_boxplot() + xlab("n_credits")

  1. From the age variable, we see that the median value for bad records is lesser than that of good records, it might be premature to say young people tend to have bad credit records, but we can safely assume it tends to be riskier.
  2. The installment_rate variable has a great deal of difference between the good and bad records, we see that bad records have almost the double median value than good ones.
  3. The median value and the range of the duration variables appears to be on the higher side of bad records as compared to good records
  4. For the amount variable, we observe that the amount for bad records is larger in general as compared to good ones
  5. We further built on this by plotting the density curve along the vertical line for their mean value and find that there is a great deal of difference for the duration as well as amount variable.

EDA of Categorical Data

ggplot(german_credit, aes(chk_acct, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(credit_his, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(purpose, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(saving_acct, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(other_debtor, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(sex, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(other_install, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge") 

ggplot(german_credit, aes(foreign, ..count..)) + 
  geom_bar(aes(fill = response), position = "dodge")

  1. For chk_acct we see that, the current status of the checking account matters as the frequency of the response variables is seen to differ from one sub category to another, overall A11 houses more number of bad credit records and A14 the least
  2. For credit_his, we observe that proportion of the response variable varies significantly, for categories A30, A31 we see the number of bad credit records are greater.
  3. For the purpose variable, we observe that the proportion of good and bad credit record varies also overall A44, A45, A410 and A46 seem to include more risky records.
  4. We also observe these trends in other variables like sex, other_debtor, saving_acct, other_install and foreign. Overall, the trend looks significant in saving_acct, purpose, credit_his and chk_acct as compared to others.