A Dynamic Ensemble Framework for Mining Textual Streams with Class Imbalance

Algorithm 2

Sample method.

Output: The optimal CF

Input:   𝐷_𝑡: The data chunk at each time stamp

    𝑀: The number of CTs

    𝐿: the prediction of 𝐱

    𝐸_𝑡: Misclassified subset at the 𝑡-th time stamp

    𝑅_𝑡: Rare-class subset at the 𝑡-th time stamp.

    𝑦_𝑅: The label of rare-class

 (1) At the 𝑡-th time stamp:

 (2) Obtain the 𝑓^𝐸 by Algorithm 1.

 (3) For all 𝑥_𝑘𝑗∈𝐷_𝑘, where 𝐷_𝑘 is the training chunk of 𝑓_𝑘𝑓_𝑘∈𝑓^𝐸;

 (4) Obtain the label 𝐿_𝑘𝑗 by 𝑓^𝐸

 (5) If 𝐿_𝑘𝑗≠𝑦_𝑘𝑗:

 (6) 𝐸_𝑡←𝐸_𝑡∪𝑥_𝑘𝑗;

 (7) Endif

 (8) If 𝑦_𝑘𝑗∈𝑦_𝑅

 (9) 𝑅_𝑡←𝑅_𝑡∪𝑥_𝑘𝑗;

 (10) Endif.

 (11) Endfor.

 (12) if 𝑝^𝑅_𝑡<𝑝^𝑅_𝜃:

 (13) 𝐷_𝑡← RESAMPLE𝐷_𝑡 at the percentage (1−𝑝^𝑅_𝜃)

 (14) 𝐷^train_𝑡←𝐷_𝑡∪𝐸_𝑡∪𝑅_𝑡;

 (15) Else

 (16) 𝐷^train_𝑡←𝐷_𝑡∪𝐸_𝑡∪𝑅_𝑡;

 (17) Endif.

 (18) Create a new CT 𝑓_𝑡 using 𝐷^train_𝑡 relying on Algorithm 1;

 (19) UPDATE(𝑓^𝐸)