Multi-label text document classification is more challenging than multi-class classification because it requires simultaneously assigning many class labels to unlabeled text documents rather than just one. The development of an efficient method for multi-label classification is continually in demand. On the other hand, the fuzzy rough set (FRS) is an effective mathematical tool for handling uncertain data, and it has numerous applications in classification, dimensionality reduction, and feature selection. The Fuzzy Rough Set Based on Robust Nearest Neighbor (FRS-RNN) approach has been demonstrated in the literature to be a proper classification tool for multi-class classification on both real-valued and text datasets. Due to a lack of study on FRS-RNN to demonstrate its potential power on multilabel text datasets, this research paper proposed ML-FRSRNN: a new multi-label algorithm based on FRS-RNN. In general, the process of classifying text documents has two crucial steps: extracting features and building a classifier model. For efficient feature extraction, techniques based on TF-IDF and CNN are mainly used. The CNN has the best feature engineering because it pre-processes documents well and uses pre-trained word embedding to make them easier to understand. Utilizing the CNN structure for feature extraction, the proposed ML-FRSRNN has been effectively applied to classify multi-label text documents. The classification performance of the proposed method is evaluated and compared with state-of-the-art multi-label machine learning (ML) models such as Binary Relevance (BR), Label Power Set (LP), Classifier Chains (CC), and ML-KNN using well-defined multi-label classification metrics like hamming loss, recall, precision, subset accuracy, and F1 score. The proposed ML-FRSRNN outperforms the aforementioned multilabel ML models according to the experimental findings and empirical evaluation

Multi-label text Classification, Convolutional Neural Network, Fuzzy Rough Set, Natural Language Processing, Machine Learning