Summary: | With the technological advancements, vast amounts of data are generated in various domains, leading to massive growth in data sizes, dimensions and complexity. Big data, often associated with high dimensionality, poses significant challenges to the classification process. Certain technique such as dimension reduction is required to enhance the accuracy of classification task whilst improving the prediction performance. Hence, this review aims at presenting the study of dimension reduction techniques for the classification of high-dimensional data. The Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) criteria are followed while conducting the systematic literature review. Publications from 2017 to 2021 on Scopus and Web of Science databases were searched with dimension reduction, classification and high-dimensional data as the keyword. From 129 articles, 29 studies were covered in this systematic review. In general, these studies report the performance of classification and the empirical comparisons of dimension reduction techniques applied to the high-dimensional dataset. Based on the articles reviewed, most of the proposed method significantly increases the accuracy rates and are able to improve the performance of high-dimensional data. Some of the new methods proposed resulted in higher accuracy when compared with the widely applied dimension reduction methods. Despite that, the overall evaluation showed that all studies presented the efficacy of the proposed dimension reduction approaches along with the improvement in the classification performance. However, there is no clear winner which dimensionality reduction technique performs best but reducing the number of features when analyzing the high-dimensional data is significant for the accuracy performance. © 2023 Author(s).
|