Skip to content

README.md

Latest commit

 

History

History
57 lines (42 loc) · 2.08 KB

README.md

File metadata and controls

57 lines (42 loc) · 2.08 KB
Raw

ExoplanetAI

Website: https://exoplanetai.onrender.com/

Upload

Problem

Currently, much of exoplanet candidate vetting is performed manually by NASA scientists and researchers. That manual review is time-consuming and doesn't scale well as survey missions collect ever larger volumes of data. This project aims to help by automating the first-pass classification of candidates so experts can focus on the most promising leads.

Challenge objective

  • Train an ML model (using Kepler/K2/TESS data) to classify exoplanet candidates as CONFIRMED, CANDIDATE, or FALSE.
  • Provide a simple web interface to upload CSV data and see predictions and confidence scores.
  • Use reproducible preprocessing (handle missing values, save preprocessing artifacts) so inference matches training.
  • Make results downloadable and expose a small API for programmatic use.

Required CSV header:

Required CSV

Required columns

Required CSV header:

koi_score,koi_depth,koi_model_snr,koi_period,koi_duration,koi_prad,koi_srad,koi_kepmag,koi_teq

Example row:

1.0,1288.3,87.2,9.27358173,3.2875,2.47,0.696,15.302,649.0

How it works (simple)

  • Train once: model_trainer.py learns from past labeled data and saves the model in models/.
  • Webapp loads the saved model (now in background) and serves the upload UI.
  • When you upload a CSV the app:
    1. Checks the required columns are present.
    2. Fills any missing numbers with column medians.
    3. Sends rows to the RandomForest model which combines many decision trees.
  • The app shows:
    • Prediction label (CONFIRMED / CANDIDATE / FALSE)
    • Confidence (top probability, e.g. 96.7%)
    • Per-class probabilities (they add up to 100%)

Example output for the sample row above:

Prediction: CANDIDATE
Confidence: 96%
Probabilities: {CANDIDATE: 0.96, CONFIRMED: 0.03, FALSE: 0.01}

Result

Results

Output

Output: Prediction label, Confidence (%), per-class probabilities.