Generalized k-labelset Ensemble for multi label classification and cost-sensitive classification: Review

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 and uses Label powerset (LP) classifier. RAkEL assumes equal weightage for each label set. To overcome this drawback, a new approach as reported in the literature that is Generalized k-labelset ensemble (GLE) advocates the basis expansion model to train LP classifier on random k label set. To reduce the global error between the estimate and ground truth, the expansion coefficients are learned. This model is further extended to solve the multi label misclassification problems. It is capable of handling noisy data sets such as social tagging by treating tag count as misclassification cost. This paper gives the review of GLE method.