| Adjusted Jaccard Distance (AJD) |
 |
|
[0, 1] |
0 |
Quantifies how much local neighborhood structure is preserved after dimensionality reduction. |
| co-KNN (AUC & size) |
|
 |
– |
The higher the better |
Evaluates how well local neighborhood structures are preserved. |
| Continuity |
|
|
[0, 1] |
1 |
Measures how many high-dim neighbors are preserved in low dimensions. |
| Distance correlation |
 |
|
[0, 1] |
1 |
Detects linear and nonlinear dependencies; equals 0 only when variables are independent. |
| Density Preservation |
|
|
[-1, 1] |
1 |
Measures how well local point densities are maintained after dimensionality reduction. |
| Entourage |
|
|
[0, 1] |
1 |
Quantifies local neighborhood preservation quality (structural fidelity). |
| Graph Edit Distance (GED) |
|
|
– |
The lower the better |
Quantifies graph dissimilarity via minimum edits between two neighborhood graphs. |
| Kruskal's Stress |
|
|
[0, 1] |
0 |
Based on the difference between distances in high and low dimensions. |
| LCMC |
|
|
– |
The higher the better |
Evaluates changes in the nearest neighbor matrix after dimension reduction. |
| Local & Global co-KNN |
 |
|
[0, 1] |
1 |
Measures preservation of both local and global neighborhood structure. |
| Spearman's Rho |
|
|
[0, ∞) |
0 |
Based on the difference in rank-ordering of distances between high and low dimensions. |
| Trustworthiness |
|
|
[0, 1] |
1 |
Measures how well low-dimensional neighbors match those in the high-dimensional space. |
| Trustworthiness and Continuity |
|
|
[0, 1] |
1 |
Combines trustworthiness and continuity scores for neighborhood structure conservation. |