Combination of Cluster Analysis with Dimensionality Reduction Techniques for Pattern Recognition Studies in Healthcare Data: Comparing PCA, t-SNE and UMAP

This study compares the performance of three dimensionality reduction techniques: Principal Component Analysis (PCA), t-Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP), when applied with a hierarchical cluster analysis for pattern recognition. We used this methodology in 240 synthetic datasets, each simulating a sample of patients with j binary diagnosis grouped in $k$ predefined clusters. For each dataset, we retrieved $k$ using hierarchical cluster analyses, and evaluated the quality and accuracy of the model using four cluster validation indices. Results suggest that UMAP performs better than PCA and is more efficient than t-SNE under these conditions.