Multi-label learning has been a topic of research interest in multimedia, text speech recognitions, music, image processing, information retrieval etc. In Multi-label classification (MLC) each instance is associated with a set of multiple class labels. Like other machine learning algorithms, data preprocessing plays an key role in MLC. Feature selection is an important preprocessing step in MLC, due to high dimensionality of datasets and associated computational costs.Extracting the most informative features considerably reduces the computational loads of MLC. Most of the Multi-label feature selection algorithms available in literature involve conversions to multiple single labeled feature selection problems. We proposed an efficient modification of a recent multi-label feature selection algorithm [1] available in literature. Our algorithm consists of two steps: in the first step we decompose the output label space into lower dimensions using simple matrix factorization method; subsequently we employ feature selection process in the decoupled reduced space. Our simulations with real world datasets reveal the efficiency of proposed framework. © 2019 IEEE.