StarTrack is an interactive web-based dashboard and machine learning pipeline for stellar classification using spectral data. It allows users to query stars by PLATE-MJD-FIBERID and view both model predictions and actual classifications. Designed as a learning experience and showcase project for me, StarTrack combines tabular and spectral analysis through a multimodal deep learning system.
You can download StarTrack's CSV data here: https://drive.google.com/file/d/170lCU-O4Cxb-RVoX_OIWrBycr3kPjQbi/view?usp=sharing
- Star Lookup by PLATE, MJD, and FIBERID
- Dual-model predictions:
- Dense Model for tabular metadata (e.g., redshift, magnitudes)
- BiLSTM Model for raw spectral sequence data
- Fusion architecture combining both representations
- Comparison of model outputs vs. actual SDSS subclass labels
- Fully responsive web dashboard built with React.js
StarTrack’s pipeline includes:
- Tabular Branch: A dense neural network trained on features like redshift, magnitude, and signal-to-noise ratio.
- Spectral Branch: A Bidirectional LSTM trained on normalized 1D spectral flux values.
- Fusion Module: Outputs from both branches are concatenated and passed through a multi-head attention mechanism to produce the final subclass prediction.
Models trained and evaluated using data from the Sloan Digital Sky Survey (SDSS)
- Source: SDSS DR17 Spectroscopic Archive
- Size: ~800,000 for dense/tabular model, ~25,000 labeled samples for LSTM & Fusion model,
- Labels: Stellar classes (e.g.,
A,M,WD)
Frontend:
- React.js
- TailwindCSS
- React Recharts
- Lucide React (For icons)
Backend / ML:
- Python
- PyTorch
- NumPy, Pandas
- Matplotlib
- Sklearn
- Node.js
- Jupyter Notebook
