Review of Multimodal Data Fusion in Machine Learning

In many fields, data on the same phenomenon can be derived from a variety of sources, including different detectors, different conditions, different experiments, and different subjects. In this article, we refer to every one of the acquisitions mentioned above frames as a "modality." It is quite uncommon for one method to provide a thorough comprehension of the issue under research, considering the diversity of natural events. There are concerns beyond employing each modality individually when a growing number of techniques can provide information on an identical framework. It is generating further levels of flexibility. Many of these issues and "challenges," as we describe them, are cross-disciplinary. The two main topics that this article addresses are "The reason why data combination is required" and "In what ways can one accomplish this?" Inspired by different instances in the areas of science and technology, the primary issue is followed by a structure based on mathematics exhibiting some of the benefits of data fusion. We address the second concern by defining "diversity" and applying many methods based on data that make use of tensor as well as matrix subdivisions. These approaches demonstrate their ability to account for variability across databases. This article aims to provide readers with an overview of the scope, possibilities, and promise of the topic, regardless of their prior knowledge.