Dealing with high dimensional multi-view data: A comprehensive review of non-negative matrix factorization approaches in data mining and machine learning

Abstract

Non-negative matrix factorization (NMF) has become a well-known model in data mining in recent years. NMF is an unsupervised algorithm that efficiently reduces the number of features while maintaining the crucial information needed to reconstruct the original data by projecting the data onto a lower-dimensional space. NMF's main goal is to automatically extract hidden patterns from high-dimensional vectors; it has been effectively used for prediction, clustering, and dimensionality reduction. On the other hand, a major problem in data mining and machine learning is multi-view decision-making. Multi-view learning is a significant issue in today's multi-modal decision-making environment since it makes use of many and frequently high-dimensional data representations to improve learning outcomes. This study aims to review the state-of-the-art NMF methods for multi-view data processing, encompassing principles, representation approaches, clustering models, and algorithms with various generalizations, developments, and modifications. The review includes discussions on different aspects of the algorithms and provides a comprehensive comparison of their advantages. Additionally, it addresses several open issues and remaining challenges. Ultimately, this review seeks to establish a framework for the NMF concept that may benefit future research.