Clairvoyant: AdaBoost with Cost-Enabled Cost-Sensitive Classifier for Customer Churn Prediction

Table 1

Comparison of cost-sensitive boosting algorithms.

Classifiers	Initial weight distribution	Is class-dependent cost-sensitive?	Reweighting
Boosting	1/N	No	No
Cost-Boost	1/N	No	Yes#
Uboost	(Unequal)^$	No	No
Cost-Uboost	(Unequal)^$	No	Yes^#
AdaBoost	1/N (Equal)	No	No
AdaCost	1/N (Equal)	Yes	Yes
	Equal	No	Yes^#
	Equal	No	Yes^#
	Equal	No	Yes^#
	Equal	No	Yes^#
	Equal	No	Yes^#

$ = initial weights: , : updated new weight = cost of misclassification; old weight (if wrongly classified); new weight = old weight (if correctly classified); # = algorithm specific weight update equation.