Team: WarpDrive
| Document | Description |
|---|---|
| Research Document | Full methodology, literature review, architecture |
| Prototype Folder | Files And Datasets Of Prototype |
| Prototype Notebook | Working ML model on real xCTL v8.01 data |
| PPT Structure | Slide-by-slide content guide |
AI-enabled Detection of Exoplanets from Noisy Astronomical Light Curves
- Detect periodic transit signals in TESS data
- Classify: planet candidate / eclipsing binary / stellar blend / starspot
- Estimate parameters: period, duration, depth with uncertainty
| Stage | Tool | Purpose |
|---|---|---|
| Data | lightkurve + xCTL v8.01 |
TESS light curves + 9.5M catalog records |
| Detrend | wotan + CBVCorrector |
Remove systematics, preserve transits |
| Search | transitleastsquares |
93% detection vs 76% BLS |
| Classify | XGBoost + RF + GB ensemble | 46 physics-informed features |
| Fit | batman + emcee MCMC |
Transit parameters + uncertainties |
| Output | SNR + confidence + plots | Publication-ready results |
Key innovation: Uses contratio (contamination ratio) from TIC v8.1 as ML feature — directly addresses crowded-field blending.
- Dataset: Real 497MB xCTL v8.01 (9.5M records)
- Processing: Chunk-based streaming (500k-row blocks)
- Baseline: XGBoost — 72% accuracy
- Classes:
planetcandidate/cooldwarfs_v8/hotsubdwarfs_v8 - Challenge: Extreme imbalance —
hotsubdwarfs_v8= 0.03%
| Role | Name | College |
|---|---|---|
| Team Lead | Omprakash Behera | GCE Kalahandi |
| Member 1 | Anish Prasad | GCE Kalahandi |
| Member 2 | Apurva Nathrao Sakhare | symbiosis international University Pune |
| Member 3 | Sujoy Bose | GCE Kalahandi |
- Hackathon: ISRO BAH 2026
- Problem: PS-07
- Status: Prototype validated on real TESS data