Efficient Multi-label Classification using Attribute and Instance Selection
Shirish S. Sane · Bioscience Biotechnology Research Communications · 2020
Both attribute and instance selection is proven to be beneficial to reduce the computational complexity of classifiers while improving their accuracy.Instances in multi-label data are associated with multiple labels.Hence the process of attribute selection from multi-label data is different as compared to single-label classification.Either transformation or adaptation approaches are used by various researchers while performing attribute selection.In this paper, attribute selection and sampling are performed on the multi-label data.This pre-processed multi-label data is then fed to the proposed algorithms, namely MLFLD and its extension MLFLD-MAXP.An empirical evaluation is carried out to study the behaviour of proposed multi-label classifiers.The methods used in this work are defined as algorithms MLFS, MLIS, and MLFSIS.Comparing proposed algorithms with and without MLFS, MLIS, and MLFSIS has shown the effectiveness of using only sampling, or attribute selection followed by sampling on multi-label data.Attribute and instance selection together are noticed to be very useful for the performance enhancement of proposed algorithms over only attribute or instance selection.