차원축소

Feature 개수가 너무 적다 → 모델 성능 저하

Feature 개수가 너무 많다 → Overfit 발생

Method

1. Feature Elimination
2. Feature Selection
3. Feature Extraction
   1. Linear Projection a.k.a. Matrix Factorization
      1. PCA (Principal Component Analysis)
      2. SVD (Singular Value Decomposition)
      3. Factor Analysis
      4. NMF (Non-negative Matrix Factorization)
      5. LDA (Linear Discriminant Analysis)
   2. Non-Linear Manifold Learning a.k.a. Neighbor Graph
      1. t-SNE
      2. UMAP
      3. LLE (Locally-Linear Embedding)
      4. Isomap
      5. Kernel PCA
      6. Autoencoder
      7. SOM (Self -Organizing Map)
      8. GDA (Generalized Discriminant Analysis)

Reference

차원 축소 (Dimensionality Reduction)

차원 축소 알고리즘을 비교해보자 (PCA, T-sne, UMAP)

차원축소 tsne, pca와 비교

Dimensionality reduction