R/build_iv_table.R

In louis-vines/miv.select: A package for aided manual variable selection in binary logistic regression based models using marginal information value of features.

build_iv_table <- function(dframe, y){
  y_sym = as.symbol(y)

  total_goods <- sum(dframe[[y]] == 0)
  total_bads <- sum(dframe[[y]] == 1)
  total_log_odds <- log(total_bads / total_goods)

  iv_table <- dframe %>%
    group_by(group) %>%
    summarise(freq = n(),
              freq_bad = sum((!!y_sym) == 1),
              freq_good = sum((!!y_sym) == 0)) %>%
    mutate(prop_total = freq / sum(freq),
           bad_rate = freq_bad / freq,
           odds = freq_bad / freq_good,
           log_odds = log(odds),
           woe = log_odds - total_log_odds,
           iv = (freq_bad/total_bads - freq_good/total_goods) * woe) %>%
    mutate_if(is.numeric, ~ round(.x, 5)) %>%
    mutate(group = forcats::fct_explicit_na(group))

  attr(iv_table$group, "class") <- "factor"

  iv_table
}