Research Report
A shareable summary of your ML pipeline.
Exoplanet Candidate Classification
Prepared by ExoExplain AI research pipeline
01. Project Overview
ExoExplain AI classifies NASA exoplanet candidates using ensemble ML models with SHAP-based explainability.
02. EDA Summary
9,564 objects across 48 features. Class balance skewed toward Confirmed (50.4%).
03. Data Cleaning
Imputed 1.3% missing values with median/mode. Winsorized outliers at 1st and 99th percentiles.
04. Model Comparison
XGBoost delivered the top F1 (0.980), followed by LightGBM (0.978) and CatBoost (0.975).
05. Performance Metrics
Accuracy 98.4% · Precision 97.9% · Recall 98.2% · F1 98.0% · ROC-AUC 0.987.
06. Confusion Matrix
True positives dominate the diagonal; only 62 confirmed objects misclassified as candidates.
07. ROC Curve
AUC of 0.987 indicates near-perfect separability between classes.
08. Feature Importance
Transit Depth, Orbital Period, and SNR are the three most influential features.
09. Prediction Explanation
SHAP values reveal how each feature pushed the current prediction above or below the base rate.
10. Conclusion
The pipeline delivers state-of-the-art accuracy with human-readable justifications suitable for astronomers.