🪐 Exoplanet Classifier 2.0
Model repo: mibrahimzia/exoplanet_model2.0
Model Description:
This model classifies potential exoplanet detections as:
✅ Confirmed Exoplanet
🟡 Candidate
❌ False Positive
It is trained on NASA’s Kepler (KOI cumulative) and TESS (TOI) catalogs, preprocessed into a tabular dataset of orbital and stellar parameters.
Algorithm: XGBoost classifier
Version: 2.0 (merged Kepler + TESS, larger dataset)
Input features:
orbital_period (days)
transit_duration (hours)
planet_radius (Earth radii)
star_temp (Kelvin)
star_radius (Solar radii)
Labels:
1 = Confirmed
0 = Candidate
-1 = False Positive
The model demonstrates strong performance in distinguishing confirmed exoplanets from candidates/false positives:
Accuracy: ~91% (on validation set)
ROC-AUC: 0.93 (see ROC curve below)
Precision/Recall:
Confirmed (Label 1): Precision ~0.90, Recall ~0.88
Not Confirmed (Labels 0/-1): Precision ~0.92, Recall ~0.94
ROC Curve
The ROC curve shows the model achieves an AUC of 0.93, indicating excellent separability.
(insert your ROC curve plot here)
Probability Distribution
The predicted probabilities clearly separate confirmed planets (green) and non-confirmed (red). The default threshold of 0.5 provides a good trade-off between precision and recall.
