groups.md

Clustering

Notes from clustering experiments

• myopia-pca-kMeans.ipynb
  Unbalanced data

  • PCA components: 10 (90+% Variance Explained)
  • KMeans Clusters: 3 is better than 4 or 5
  • Tested KNN on first cluster
    • With K: 3 and oversampling, Cluster 1:
    • Accuracy: 64%
    • Precision: 12%
    • Recall: 22% Note: A noticable decrease in accuracy compared to KNN with Balanced Data (see above).

  Balanced data with over sampling

  • PCA components:

• myopia-pca-kMeans-KNN.ipynb

• myopia-pca-tSNE.ipynb