Cost-Sensitive Random Pair Encoding for Cost-Sensitive Multi-Label Classification

Yao-Yuan Yang, Chih-Wei Chang, Hsuan-Tien Lin

Basic Idea 

OVO Hsuan-Tien Lin's MOOC handout http://www.csie.ntu.edu.tw/~htlin/mooc/doc/11_present.pdf

Cost-Sensitive Random-Pair Encoding (CSRPE)

nearest-neighbor

Experiment converge

f1 bibtex (K=499)	f1 scene (K=294)

Comparisons (Experiments)

CSRPE versus other competitors based on t-test at 95% confident level (win/tie/loss)

criteria	BR	RAKEL	HAMR	CFT
hamming	2/4/4	4/3/3	2/4/4	5/1/2
f1	9/1/0	9/1/0	9/1/0	8/0/0
rankloss	7/2/1	9/0/1	7/2/1	4/1/3
accuracy	9/0/1	9/1/0	8/2/0	8/0/0
total	27/7/6	31/5/4	26/9/5	25/2/5

Speed (v.s. CFT)

training time	prediction time

Thank you! Any question?