Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/LauraSilRu/exoplanet-profiler/llms.txt

Use this file to discover all available pages before exploring further.

ExoProfiler evaluates clustering quality using two complementary metrics — the Elbow method (inertia/WCSS) and the Silhouette coefficient — across K values from 2 to 10. K=2 is selected based on a clear Silhouette peak, and its stability is confirmed through multi-seed perturbation and cross-validation with alternative algorithms.

Why These Metrics

  • Elbow method (Inertia/WCSS): Measures total within-cluster sum of squared distances. The optimal K is at the “elbow” where adding more clusters yields diminishing inertia reduction.
  • Silhouette coefficient: Measures how similar each point is to its own cluster versus the nearest cluster. Range: [-1, 1]. Higher = better separation. Cannot be computed for K=1.

K Sweep Results

KInertiaSilhouette
24023.360.4950
32820.810.3223
42392.980.3238
5~2100< 0.28
6–10< 0.28
K=2 achieves the highest Silhouette score (0.4950), indicating strong cluster cohesion and separation. K=3 and K=4 show nearly identical Silhouette scores (~0.32), suggesting no natural three- or four-family structure in the data.

Final Model Parameters

from sklearn.cluster import KMeans

kmeans = KMeans(
    n_clusters=2,
    random_state=42,
    n_init=10  # sklearn default
)
kmeans.fit(X_pca)  # X_pca: 731 × 4 PCA component scores

Stability Analysis

To confirm that the K=2 solution is not an artifact of centroid initialization, the model is re-run across five different random seeds.
SeedInertiaSilhouetteAgreement
424023.360.4950baseline
104023.360.4950100.00%
20264023.360.4950100.00%
99994023.360.4950100.00%
1004025.210.479699.04%
4 out of 5 seeds produce identical solutions. Seed 100 differs in 7 assignments (0.96% of 731 planets), with a minor inertia increase of 1.85.

Cross-Validation with Alternative Algorithms

Hierarchical (Agglomerative) Clustering

  • Algorithm: Ward linkage, n_clusters=2
  • Agreement with K-Means: 97.13%
  • The dendrogram displays two clearly separable primary branches at the top level.

DBSCAN (Density-Based)

  • Independently identifies one dense core cluster (majority) and ~60 outlier points (label = -1).
  • These outliers correspond closely to Cluster 1 (Exotic Objects) from K-Means.
  • Confirms that the Exotic group represents a statistically sparse region of the feature space.

Decision Summary

K=2 is selected as the primary segmentation. It achieves the highest Silhouette score (0.4950), shows 99%–100% stability across 5 random seeds, and is confirmed by both hierarchical clustering (97.13% agreement) and DBSCAN outlier detection.

Interpreting Silhouette Scores

Silhouette scores can be interpreted against the following general benchmarks:
RangeInterpretation
0.71–1.00Strong structure
0.51–0.70Reasonable structure
0.26–0.50Weak structure (may be artificial)
< 0.25No substantial structure
ExoProfiler’s score of 0.4950 approaches the reasonable-structure range, indicating meaningful but not perfectly separated clusters — consistent with a continuous physical distribution rather than discrete populations.

Build docs developers (and LLMs) love