Review of High-Dimensional and Complex Data Visualization
Paska Marto Hasugian, Herman Mawengkang, Poltak Sihombing, Syahril Efendi · 2023
In the age of big data, high-dimensional and complex data create significant challenges for data visualization. This research analyzes their impact on visualization techniques. Increasing data dimensions make traditional methods struggle to depict complex structures and patterns. This study comprehensively explores challenges arising from high dimensionality and data complexity in visualization. By examining various techniques and their suitability for such data, we aim to reveal the pros and cons of each approach in conveying essential information. Additionally, this research investigates using dimension reduction techniques to tackle the complexities of the data. The effectiveness of these techniques in preserving crucial information while enhancing visualization quality is evaluated. Some of the models examined include Multidimensional Scaling (MDS), which projects data into a lower-dimensional space to maintain distance relationships but does not consider class labels. MDS operates with a distance matrix but is sensitive to data that do not conform to distance metrics and can yield suboptimal projections in high-dimensional space. Principal Component Analysis (PCA) focuses on dimension reduction by maximizing data variance but does not consider class labels. PCA requires high-dimensional data and cannot handle non-linear structures. Linear Discriminant Analysis (LDA) is used for classification by maximizing inter-class distances, minimizing intra-class distances, and requiring class labels. However, LDA works well only under the assumptions of normal distribution and fulfilled class variance differences. Conversely, t-SNE and UMAP reveal data cluster structures. t-SNE is computationally complex and sensitive to parameters, while UMAP performs faster. t-SNE interprets data with topological and network context, suited for data with strong topological structures but requiring deep domain knowledge. Each method has its uses and limitations, depending on analysis goals and data nature. This research offers valuable insights into how complex, high-dimensional data impacts visualization. Understanding this aids practitioners and researchers in selecting appropriate visualization techniques, enhancing interpretation and insights. Ultimately, it contributes to effective visualization strategies for modern, high-dimensional, and complex datasets