Fyzzy rule classifier for generalized k labelset ensemble

In multi-label classification, set of labels are associated with each example. An algorithm called Random k-labelsets (RAkEL) is an algorithm for multi-label classification that follows problem transformation approach. RAkEL algorithm uses Label powerset (LP) classifier and it assumes equal weightage for each label set. To overcome this drawback, a new approach is reported in the literature that is GLE. GLE performs the basis expansion method to train LP classifier on random k labelsets. To decrease the global error between the estimated and ground truth, the expansion coefficients are learned. GLE uses SVM classifier which uses crisp vales as the base classifier. Fuzzy rule classifier (FURIA) as reported in literature gives the better results compared with other rule based classifiers, for problem transformation methods such as Binary Relevance, Classifier Chain, and LP. It would be interesting to observe the performance of GLE with FURIA. This work aims at implementation of GLE with FURIA algorithm and compares its performance using SVM as a base classifier. Experimental results shows that GLE using fuzzy rule classifier FURIA provides better performance in terms of hamming loss, ranking loss, subset 0/1 loss, one error, average precision.