Idea
Cloud-hosted app for automated vetting and validation of TESS exoplanet candidates using machine learning and Bayesian analysis
Research Paper
Core Innovation
This paper introduces RAVEN, a pipeline combining Gradient Boosted Decision Trees and Gaussian Process classifiers within a Bayesian framework to accurately validate exoplanet candidates. It uniquely integrates synthetic and real false positive training sets to improve vetting accuracy and scalability. The approach achieves high precision and recall on TESS data, enabling automated, statistically robust candidate validation.
Market Size (TAM)
$2–10B TAM for astronomical data analysis platforms; $1–2B SAM from space agencies and research institutions. Driven by increasing volume of exoplanet data and demand for automated vetting tools.
Potential Customers & Pain Points
- Astronomers needing efficient exoplanet candidate validation
- Space agencies managing large TESS datasets
- Research institutions lacking scalable vetting tools
Business Model
Subscription-based cloud platform with tiered access for research institutions and space agencies; API access for integration with existing tools
Competitive Landscape
- ExoMiner
- VESPA
- TRICERATOPS
Implementation Challenges
- Dependence on quality and representativeness of training data
- Integration with existing astronomical data pipelines
- Adoption by traditional astronomy research workflows
Validation Strategy
- Deploy RAVEN on additional TESS candidate datasets for real-world testing
- Collaborate with astronomy groups for feedback and iterative improvement
- Benchmark against existing vetting tools on independent datasets
Research Paper Overview
RAVEN: RAnking and Validation of ExoplaNets
Summary
We present RAVEN, a newly developed vetting and validation pipeline for TESS exoplanet candidates. The pipeline employs a Bayesian framework to derive the posterior probability of a candidate being a planet against a set of False Positive (FP) scenarios, through the use of a Gradient Boosted Decision Tree and a Gaussian Process classifier, trained on comprehensive synthetic training sets of simulated planets and 8 astrophysical FP scenarios injected into TESS lightcurves. These training sets allow large scale candidate vetting and performance verification against individual FP scenarios. A Non-Simulated FP training set consisting of real TESS candidates caused primarily by stellar variability and systematic noise is also included. The machine learning derived probabilities are combined with scenario specific prior probabilities, including the candidates' positional probabilities, to compute the final posterior probabilities. Candidates with a planetary posterior probability greater than 99% against each FP scenario and whose implied planetary radius is less than 8$R_{∎}$ are considered to be statistically validated by the pipeline. In this first version, the pipeline has been developed for candidates with a lightcurve released from the TESS Science Processing Operations Centre, an orbital period between 0.5 and 16 days and a transit depth greater than 300ppm. The pipeline obtained area-under-curve (AUC) scores > 97% on all FP scenarios and > 99% on all but one. Testing on an independent external sample of 1361 pre-classified TOIs, the pipeline achieved an overall accuracy of 91%, demonstrating its effectiveness for automated ranking of TESS candidates. For a probability threshold of 0.9 the pipeline reached a precision of 97% with a recall score of 66% on these TOIs. The RAVEN pipeline is publicly released as a cloud-hosted app, making it easily accessible to the community.