decision.tree: Custom decision tree algorithm

In PDtoolkit: Collection of Tools for PD Rating Model Development and Validation

Custom decision tree algorithm

Description

decision.tree runs customized decision tree algorithm. Customization refers to minimum percentage of observations and defaults in each node, maximum tree depth, monotonicity condition at each splitting node and statistical test (test of two proportions) used for node splitting.

Usage

decision.tree(
  db,
  rf,
  target,
  min.pct.obs = 0.05,
  min.avg.rate = 0.01,
  p.value = 0.5,
  max.depth = NA,
  monotonicity
)

Arguments

db	Data frame of risk factors and target variable supplied for interaction extraction.
rf	Character vector of risk factor names on which decision tree is run.
target	Name of target variable (default indicator 0/1) within db argument.
min.pct.obs	Minimum percentage of observation in each leaf. Default is 0.05 or 30 observations.
min.avg.rate	Minimum percentage of defaults in each leaf. Default is 0.01 or 1 default case.
p.value	Significance level of test of two proportions for splitting criteria. Default is 0.05.
max.depth	Maximum tree depth.
monotonicity	Logical indicator. If TRUE, observed trend between risk factor and target will be preserved in splitting node.

Value

The command decision.tree returns a object of class cdt. For details on output elements see the Examples.

See Also

predict.cdt

Examples

suppressMessages(library(PDtoolkit))
data(loans)
#modify risk factors in order to show how the function works with missing values
loans$"Account Balance"[1:10] <- NA
loans$"Duration of Credit (month)"[c(13, 15)] <- NA
tree.res <- decision.tree(db = loans,
	rf = c("Account Balance", "Duration of Credit (month)"), 
	target = "Creditability",
	min.pct.obs = 0.05,
	min.avg.rate = 0.01,
	p.value = 0.05,
	max.depth = NA,
	monotonicity = TRUE)
str(tree.res)

PDtoolkit documentation built on Sept. 20, 2023, 9:06 a.m.