Cost-Sensitive Classification for Evolving Data Streams with Concept Drift and Class Imbalance

Algorithm 2

Cost-sensitive based data stream algorithm.

	Input: data stream S, the maximum number of based classifiers max
	Output: ensemble E with weighted classifiers {C1,……, Cmax}
(1)	E ⟵ ϕ;
(2)	for each xi ∈ S do
(3)	⟵  ∪ {xt};
(4)	apply CS-ReliefF on Bi;
(5)	if| ≥ d or change is detected then
(6)	learn a new classifier C′ on data after feature selection;
(7)	weight C′ according to equation (6);
(8)	for Ci ∈ E do
(9)	calculate the MSEij and MSEr;
(10)	update (C′) according to equation (7);
(11)	end for
(12)	if |E| ˂ max then
(13)	add C′ into E;
(14)	else
(15)	replace the worst classifier with C′;
(16)	end if
(17)	for Cj ∈ E\{C′} do
(18)	train all Cj ∈ E on Bi;
(19)	end for
(20)	end for
(21)	end.