binsmlr, as in bin similar, is a package to create bins that are assessed to be heterogenous (in terms of default risk) across a range of ordinal values. The binning can be performed automatically, interactively and manually.
The objective is to create groups of data (bins) across a range of ordinal values. The bins produced should be homogenous within and heterogenous across bins in terms of the default risk. A binning approach, in the context of data transformation, is common in different credit risk applications and in particular to credit risk portfolios having a relatively large number of defaults. A prerequisite is that each observation is assigned an ordinal value such as score or rating etc., that indicates the relative default risk as proposed by the rank ordering model. Also, a default indicator is required.
The main process is the following:
-
Create a relatively large set of initial bins.
-
Perform homogeneity tests between all adjacent bins
-
Merge the pair of bins that are the most similar in terms of proportions of state of non-default and state of default.
-
Repeat step 2 and 3 until all bins are heterogeneous or if a minimum number of bins are obtained.
The process above should result in bins that are statistically different in terms of default risk across a range of ordinal data.
Install the development version from GitHub.
# install.packages("devtools")
devtools::install_github("rrunner/binsmlr")library(binsmlr)
help("binsmlr")
# example of interactive binning
# create initial bins
bins <- create_initial_bins(bin_data, 30, "score", "default")
length(bins)
is_monotonic(bins)
# reduce bins by performing repeated homogeneity tests
new_bins <- reduce_bins(bins, min_required_bins = 7, confidence_level = 0.01)
length(new_bins)
is_monotonic(new_bins)
# plot initial and reduced bins
bins_df <- dplyr::bind_rows(bins)
plot(x = bins_df$mid_score, y = log(bins_df$odds), type = "p",
col = "lightblue", cex = 1.5, pch = 20, ylab = "log(odds)",
xlab = "score")
new_bins_df <- dplyr::bind_rows(new_bins)
points(x = new_bins_df$mid_score, y = log(new_bins_df$odds),
col = "darkblue", cex = 1.5, pch = 20)
legend(x = "topright", legend = c("Initial bins", "Reduced bins"),
col = c("lightblue", "darkblue"), pch = 20, pt.cex = 1.5, bty = "n")If you encounter a clear bug, please file an issue with a minimal reproducible example on GitHub.