Document classification is one of the important fields of text mining. At present, category identification using taxonomy for scientific publications is a manual task. These taxonomies support authors which contain a large number of classes organized in the form of hierarchy that becomes quite difficult to choose a relevant category or categories for their work. Due to the amalgamation of research work in multiple domains, the problem becomes a multi-label classification (MLC). MLC is broadly solved using two different approaches (Problem Transformation and algorithm adaptation). In literature, a lot of single label classifiers are available to deal with single label dataset such as Support Vector Machine (SVM), K Nearest Neighbour (KNN), Naive Bayes and many more, these classifiers have low accuracy on text datasets due to the similarity measures and inappropriate selection of features. Similar approaches exist, which transform the multi label dataset into binary classification problems such as Binary Relevance (BR), Classifier Chain (CC), Probabilistic Classifier Chain (PCC) and many more. These algorithms also have a very low accuracy for text data. The issue which has not given proper importance is the order in which the binary classifiers are executed. Algorithm adaptation techniques such as decision trees, SVM, Multi-label K Nearest Neighbour (ML-KNN) and neural network also exist for MLC but have low accuracy due to similar weightage of features for all labels and have never been tested for a scientific publication datasets. The algorithm adaptation approaches have never been studied with feature weighting as all the features may not play the same role for each label in the MLC. We argue that all the approaches to deal with MLC for scientific documents suffer from low accuracy.