Hunting New Exoplanets with Machine Learning
An autonomous research pipeline hunting for new exoplanets in NASA TESS data. Currently searching 19,711 unsearched M-Dwarf stars with 40 experiments and ExoMiner validation.
40
Experiments
9670.0%
Model AUC
19,711
Stars Searching
How It Works
NASA TESS Data
2,392 light curves from the Transiting Exoplanet Survey Satellite
Light Curves
Time-series brightness measurements - a planet crossing its star causes a tiny dip
Feature Extraction
24+ engineered features capturing transit shape, depth, noise, and stellar properties
AutoResearch Loop
AI agent runs 40 experiments autonomously - testing hypotheses, tuning models
Best Model
4-model ensemble (XGBoost + RF + GBM + ExtraTrees) with 96.6% AUC
4,629 Candidates
Unclassified Planet Candidates scored by planet probability
NASA TESS Data
2,392 light curves from the Transiting Exoplanet Survey Satellite
Light Curves
Time-series brightness measurements - a planet crossing its star causes a tiny dip
Feature Extraction
24+ engineered features capturing transit shape, depth, noise, and stellar properties
AutoResearch Loop
AI agent runs 40 experiments autonomously - testing hypotheses, tuning models
Best Model
4-model ensemble (XGBoost + RF + GBM + ExtraTrees) with 96.6% AUC
4,629 Candidates
Unclassified Planet Candidates scored by planet probability
The Research Journey
From first experiment to real discovery pipeline — built in 3 days of intensive autonomous research.
Mar 14
First Training
Baseline model: 0.882 AUC on 1,693 targets
Mar 14
CROWDSAP Breakthrough
FITS metadata feature adds +0.040 AUC — biggest single improvement
Mar 14
AutoResearch Loop
AI autonomously runs 40 experiments. Feature engineering plateau at 0.929 AUC
Mar 15
Extended Dataset
2,441 targets (+41% more data). AUC jumps to 0.967 — more data beats better algorithms
Mar 15
CNN Attempt
Deep learning fails at 0.625 AUC — too few samples for neural networks. XGBoost wins.
Mar 16
ExoMiner Reality Check
NASA's classifier scores our top candidates near 0. Correlation: 8.3%. Humbling but educational.
Mar 16
Physics Features
depth_vs_jupiter and duration_ratio — first features with real physical separation between planets and FPs
Mar 16
PIVOT to Discovery
From re-classifying known candidates to searching 19,711 unsearched M-Dwarf stars
Model Performance
9670.0%
Validation AUCⓘ
92.1%
Recallⓘ
87.8%
Precisionⓘ
40
Experiments
Experiment Journey
Phase 1 — Algorithm Optimization
40 experiments on 1,693 training targets. AutoResearch autonomously tested hypotheses — feature engineering, model types, hyperparameters. Hard ceiling at ~92.9% AUC. The algorithm hit its limit with the available data.
Phase 2 — Extended Dataset (+41%)
Dataset expanded to 2,392 targets via NASA S3. Same model architecture, more training data. AUC jumped to 96.6% — proving that data volume, not algorithmic complexity, was the bottleneck.
What is AUC?
A single number that captures how well the model distinguishes real planets from false positives. 1.0 = perfect, 0.5 = random coin flip. Our 96.6% means the model is highly reliable.
ExoMiner — Validated by NASA
What is ExoMiner?
ExoMiner is NASA JPL's deep learning classifier — the gold standard for transit vetting. It uses centroid shifts, pixel-level difference images, and 20+ diagnostic signals that go far beyond light curve analysis alone. It has validated 301+ new exoplanets.
Our Comparison
We ran ExoMiner on 2,228 Planet Candidates alongside our model. The correlation was just 8.3% — because ExoMiner sees spatial information we can't access. This taught us where our model's blind spots are and led to our pivot toward discovery.
How we use ExoMiner now: Every candidate from our blind M-Dwarf search will be validated by ExoMiner before submission. Our TLS pipeline finds the signal. Our ML model provides a first assessment. ExoMiner confirms or denies. Two independent systems agreeing = high confidence discovery.
The Threshold Decision
A model doesn't just say "planet" or "not planet" - it outputs a probability. The threshold is where we draw the line.
Threshold 0.50 (standard)Finds 86% of planets
Misses 14% of real planets
Threshold 0.32 (our choice)Finds 92.1% of planets
Misses only 7.9% of real planets
Why lower? In planet hunting, missing a real signal is worse than investigating a false one. A false positive costs a few hours of follow-up. A missed planet is a discovery lost forever. We tuned our model to be sensitive.
Discovery Pipeline Status
Blind Search Running
19,711
M-Dwarf Targets
TLS
Transit Search Active
50-165
Expected Candidates
The pipeline is searching 19,711 M-Dwarf stars that have never been individually analyzed for transiting planets. Using Transit Least Squares on 6 CPU cores. Results expected by morning. Each candidate will be validated by both our ML model and NASA's ExoMiner.
What's Next
🔭Running
Blind Search Results
19,711 M-Dwarf stars being searched overnight. Expected yield: 50-165 new planet candidates based on occurrence rates.
🤖Queued
ExoMiner Validation
Every candidate will be independently validated by NASA's ExoMiner classifier. Only dual-confirmed signals advance.
📡Planned
Community Submission
Validated candidates submitted as CTOIs to NASA's ExoFOP platform for follow-up by the global astronomy community.
📄Planned
Research Publication
Methodology paper describing the autonomous pipeline — publishable in AJ or MNRAS even without confirmed planets.
Glossary — what do these terms mean?
AUC
Area Under the Curve. Measures model accuracy from 0–100%. 96.6% means we correctly classify planets vs. false signals 96.6% of the time.
SDE
Signal Detection Efficiency. How clearly a transit signal stands out from noise. SDE > 50 is a strong detection, > 100 is exceptional.
TESS
Transiting Exoplanet Survey Satellite — NASA's space telescope launched 2018, surveys the sky for transiting exoplanets.
TOI
TESS Object of Interest — a star identified by NASA as potentially hosting a planet based on TESS data.
TLS
Transit Least Squares — a statistical algorithm that searches light curves for the characteristic U-shaped dip of a planetary transit.
XGBoost
A machine learning algorithm (eXtreme Gradient Boosting). Excellent at pattern recognition in structured data like our transit features.
Recall
How many real planets we find out of all real planets. 92.1% recall = we catch 92 of every 100 real signals.
Precision
How many of our 'planet' predictions are actually correct. 87.8% precision = ~88% of our flagged signals are genuine.
Transit
When a planet passes in front of its star, blocking a tiny fraction of the starlight — creating a brief, repeating dip in brightness.
Light Curve
A chart showing a star's brightness over time. The 'dip' in a light curve is the signature of a transiting planet.
False Positive
A signal that looks like a planet but isn't — could be a binary star, background noise, or instrument artifacts.
ML Score
Our model's confidence that a candidate is a real planet (0.0 to 1.0). Higher = more likely a real planet.
M-Dwarf
Small, cool red stars (2700-3600K). Ideal for finding Earth-sized planets because transits create deeper dips relative to the star's small size.
ExoMiner
NASA JPL's deep learning classifier — the gold standard for transit vetting. Uses pixel-level data and 20+ diagnostic signals.
TLS
Transit Least Squares — an algorithm that searches light curves for the characteristic U-shaped dip of a planetary transit, using a physically realistic model.
CTOI
Community TOI — planet candidates discovered by independent researchers and submitted to NASA's ExoFOP platform for follow-up.
Want to Contribute?
The code is open source - run it yourself, improve the model, hunt planets. The ML pipeline and this frontend are both on GitHub.