Optimal Binning for Scoring Modeling

The binning_by() finding intervals for numerical variable using optical binning. Optimal binning categorizes a numeric characteristic into bins for ulterior usage in scoring modeling.

binning_by(.data, y, x, p = 0.05, ordered = TRUE, labels = NULL)

Arguments

.data	a data frame.
y	character. name of binary response variable(0, 1). The variable must contain only the integers 0 and 1 as element. However, in the case of factor having two levels, it is performed while type conversion is performed in the calculation process.
x	character. name of continuous characteristic variable. At least 5 different values. and Inf is not allowed.
p	numeric. percentage of records per bin. Default 5% (0.05). This parameter only accepts values greater that 0.00 (0%) and lower than 0.50 (50%).
ordered	logical. whether to build an ordered factor or not.
labels	character. the label names to use for each of the bins.

Value

an object of "optimal_bins" class. Attributes of "optimal_bins" class is as follows.

class : "optimal_bins".
type : binning type, "optimal".
breaks : numeric. the number of intervals into which x is to be cut.
levels : character. levels of binned value.
raw : numeric. raw data, x argument value.
ivtable : data.frame. information value table.
iv : numeric. information value.
target : integer. binary response variable.

Details

This function is useful when used with the mutate/transmute function of the dplyr package. And this function is implemented using smbinning() function of smbinning package.

attributes of "optimal_bins" class

Attributes of the "optimal_bins" class that is as follows.

class : "optimal_bins".
levels : character. factor or ordered factor levels
type : character. binning method
breaks : numeric. breaks for binning
raw : numeric. before the binned the raw data
ivtable : data.frame. information value table
iv : numeric. information value
target : integer. binary response variable

See vignette("transformation") for an introduction to these concepts.

Examples

# \donttest{
library(dplyr)

# Generate data for the example
heartfailure2 <- heartfailure
heartfailure2[sample(seq(NROW(heartfailure2)), 5), "creatinine"] <- NA

# optimal binning using character
bin <- binning_by(heartfailure2, "death_event", "creatinine")
#> Warning: The factor y has been changed to a numeric vector consisting of 0 and 1.
#> 'Yes' changed to 1 (positive) and 'No' changed to 0 (negative).

# optimal binning using name
bin <- binning_by(heartfailure2, death_event, creatinine)
#> Warning: The factor y has been changed to a numeric vector consisting of 0 and 1.
#> 'Yes' changed to 1 (positive) and 'No' changed to 0 (negative).
bin
#> binned type: optimal
#> number of bins: 3
#> x
#> [0.5,0.9] (0.9,1.8] (1.8,9.4]      <NA> 
#>        80       167        47         5 

# performance table
attr(bin, "performance")
#>         Bin CntRec CntPos CntNeg CntCumPos CntCumNeg RatePos RateNeg RateCumPos
#> 1 [0.5,0.9]     80      9     71         9        71 0.09375 0.34975    0.09375
#> 2 (0.9,1.8]    167     52    115        61       186 0.54167 0.56650    0.63542
#> 3 (1.8,9.4]     47     34     13        95       199 0.35417 0.06404    0.98958
#> 4      <NA>      5      1      4        96       203 0.01042 0.01970    1.00000
#> 5     Total    299     96    203        NA        NA 1.00000 1.00000         NA
#>   RateCumNeg    Odds   LnOdds      WoE      IV     JSD     AUC
#> 1    0.34975 0.12676 -2.06546 -1.31660 0.33705 0.03933 0.01639
#> 2    0.91626 0.45217 -0.79369 -0.04483 0.00111 0.00014 0.20654
#> 3    0.98030 2.61538  0.96141  1.71027 0.49620 0.05542 0.05203
#> 4    1.00000 0.25000 -1.38629 -0.63744 0.00592 0.00073 0.01960
#> 5         NA 0.47291 -0.74886       NA 0.84028 0.09562 0.29457

# summary optimal_bins class
summary(bin)
#> ── Binning Table ──────────────────────── Several Metrics ── 
#>         Bin CntRec CntPos CntNeg RatePos RateNeg    Odds      WoE      IV
#> 1 [0.5,0.9]     80      9     71 0.09375 0.34975 0.12676 -1.31660 0.33705
#> 2 (0.9,1.8]    167     52    115 0.54167 0.56650 0.45217 -0.04483 0.00111
#> 3 (1.8,9.4]     47     34     13 0.35417 0.06404 2.61538  1.71027 0.49620
#> 4      <NA>      5      1      4 0.01042 0.01970 0.25000 -0.63744 0.00592
#> 5     Total    299     96    203 1.00000 1.00000 0.47291       NA 0.84028
#>       JSD     AUC
#> 1 0.03933 0.01639
#> 2 0.00014 0.20654
#> 3 0.05542 0.05203
#> 4 0.00073 0.01960
#> 5 0.09562 0.29457
#> 
#> ── General Metrics ───────────────────────────────────────── 
#> • Gini index                       :  -0.41087
#> • IV (Jeffrey)                     :  0.84028
#> • JS (Jensen-Shannon) Divergence   :  0.09562
#> • Kolmogorov-Smirnov Statistics    :  0.28084
#> • HHI (Herfindahl-Hirschman Index) :  0.40853
#> • HHI (normalized)                 :  0.21137
#> • Cramer's V                       :  0.41553 
#> 
#> ── Significance Tests ──────────────────── Chisquare Test ── 
#>       Bin A     Bin B statistics      p_value
#> 1 [0.5,0.9] (0.9,1.8]   11.50352 6.946445e-04
#> 2 (0.9,1.8] (1.8,9.4]   25.90425 3.587780e-07
#> 

# visualize all information for optimal_bins class
plot(bin)

# visualize WoE information for optimal_bins class
plot(bin, type = "WoE")

# visualize all information without typographic
plot(bin, typographic = FALSE)

# extract binned results
extract(bin) %>%
  head(20)
#>  [1] (1.8,9.4] (0.9,1.8] (0.9,1.8] (1.8,9.4] (1.8,9.4] (1.8,9.4] (0.9,1.8]
#>  [8] (0.9,1.8] (0.9,1.8] (1.8,9.4] (1.8,9.4] [0.5,0.9] (0.9,1.8] (0.9,1.8]
#> [15] (0.9,1.8] (0.9,1.8] [0.5,0.9] [0.5,0.9] (0.9,1.8] (1.8,9.4]
#> Levels: [0.5,0.9] < (0.9,1.8] < (1.8,9.4]
# }

Arguments

Value

Details

attributes of "optimal_bins" class

See also

Examples