# Traditional Credit Scoring Using Logistic Regression
library(data.table)
library(scorecard)
# data prepare ------
# load germancredit data
data("germancredit")
# set creditability as 1 or 0
dt = setDT(germancredit)[,creditability := ifelse(creditability=="bad", 1, 0)]
# filter variable via missing rate, iv, identical value rate
dt_s = var_filter(dt, y="creditability")
# breaking dt into train and test
dt_list = split_df(dt_s, y="creditability", ratio = 0.6, seed = 30)
train = dt_list$train; test = dt_list$test;
# woe binning ------
bins = woebin(dt_s, y="creditability", print_step = 1)
# woebin_plot(bins)
# binning adjustment
## adjust breaks interactively
# breaks_adj = woebin_adj(bins, dt_s, "creditability")
## or specify breaks manually
breaks_adj = list(
age.in.years=c(26, 35, 40),
other.debtors.or.guarantors=c("none", "co-applicant%,%guarantor"))
bins_adj = woebin(dt_s, y="creditability", breaks_list=breaks_adj, print_step=0)
# converting train and test into woe values
train_woe = woebin_ply(train, bins_adj, print_step=0)
test_woe = woebin_ply(test, bins_adj, print_step=0)
# glm ------
m1 = glm( creditability ~ ., family = "binomial", data = train_woe)
# summary(m1)
# Select a formula-based model by AIC
m_step = step(m1, direction="both", trace = FALSE)
m2 = eval(m_step$call)
# summary(m2)
# performance ks & roc ------
# predicted proability
train_pred = predict(m2, train_woe, type='response')
test_pred = predict(m2, test_woe, type='response')
# performance
train_perf = perf_eva(train$creditability, train_pred, title = "train")
test_perf = perf_eva(test$creditability, test_pred, title = "test")
# score ------
card = scorecard(bins_adj, m2)
# credit score
train_score = scorecard_ply(train, card, print_step=0)
test_score = scorecard_ply(test, card, print_step=0)
# psi
perf_psi(
score = list(train = train_score, test = test_score),
label = list(train = train$creditability, test = test$creditability),
x_limits = c(250, 700),
x_tick_break = 50
)