ExoScope.
A world away.
A signal within reach.
Exploring exoplanet candidates through machine learning, feature engineering, and a visual web interface.
Open original prototypeWhen a planet crosses in front of its star, the measured brightness can dip. Adjust the depth and noise to explore this idea. Synthetic data for explanation only; this does not run ExoScope’s classifier. About transits ↗
More data.
More possibilities to explore.
Exoplanet searches rely on interpreting astronomical observations. A signal is a starting point for investigation, not confirmation of a new world.
ExoScope’s goal is to help automate data analysis and provide an accessible platform for researchers and enthusiasts. Our prototype combines machine learning with visual exploration.
From dataset
to an explorable result.
The project brings two strands together: machine-learning experiments and a web service for interacting with data.
- INPUT / 01
Exoplanet data
Choose the default dataset or upload a custom CSV in the original prototype.
- PREPARE / 02
Feature engineering
The presentation identifies feature engineering as part of the model approach.
- MODEL / 03
Tree-based learning
Compare XGBoost, LightGBM, and Random Forest on the project’s classification task.
- EXPLORE / 04
Web interface
Search, filter, visualize, and request prediction scores through the prototype.
High-level workflow based on the project presentation and prototype interface. Exact preprocessing, model inputs, and deployment details are not specified in the presentation.
Three models.
An initial comparison.
The 2025 presentation reports the following scores. XGBoost has the highest reported F1 and ROC-AUC among these three models.
| Model | F1 score | ROC-AUC |
|---|---|---|
| XGBoost | 0.85 | 0.91 |
| LightGBM | 0.83 | 0.89 |
| Random Forest | 0.82 | 0.88 |
Source: ExoScope presentation, “Our solution.” Dataset size, split strategy, class balance, uncertainty estimates, and evaluation protocol are not specified in the presentation. These are reported prototype results, not independently validated performance claims.
A prototype.
A clear set of next questions.
The presentation outlines five future improvements. These are proposed directions, not released capabilities or dated commitments.
- 01PLANNED
Model optimization
Improve and evaluate the existing modeling approach.
- 02PLANNED
Neural-network experiments
Explore neural models alongside the current tree-based baselines.
- 03PLANNED
Automatic data updates
Investigate a workflow for refreshing the available data.
- 04PLANNED
Web-platform extension
Extend the prototype’s interface and functionality.
- 05PLANNED
Open access and collaboration
Develop opportunities for wider use and collaboration.
Built by NeoNa.
Started at Space Apps.
NASA Space Apps Challenge
Challenge: “A World Away: Hunting for Exoplanets with AI.”
ExoScope was developed by NeoNa. Participation in the challenge is part of the project’s history; it does not imply a NASA partnership or endorsement.
EXPLORE THE ORIGINAL
See the prototype.
The original hackathon web application opens separately.