2026 Volume 79 Issue 4 Pages 255-260
In recent years, while the utilization of high-dimensional data, such as multi-item dietary intake profiles, has expanded in nutritional epidemiology, effective methods for summarizing and interpreting these complex data structures are increasingly required. Clustering and dimensionality reduction play essential roles in revealing latent structures within such datasets. Clustering methods, such as partitioning-based, hierarchical, density-based, and model-based approaches, classify data on the basis of similarity. Dimensionality reduction methods, including PCA (Principal Component Analysis), t-SNE (t-Distributed Stochastic Neighbor Embedding), and UMAP (Uniform Manifold Approximation and Projection), extract essential features from high-dimensional data and provide low-dimensional representations useful for visualization and analysis. Combining dimensionality reduction with clustering enables a multifaceted understanding of latent structures in high-dimensional data, yielding valuable insights for practical applications such as the identification of meaningful dietary patterns in nutritional epidemiology. This article provides an overview of these methods and their applications in nutritional epidemiology.
