I trained a neural network to find exoplanets. Here's what actually worked.

Short summary

A developer trained a 1D convolutional neural network to classify exoplanets from NASA Kepler space telescope light curves, achieving 0.96 ROC-AUC by addressing class imbalance with weighted loss and preventing data leakage. The project includes a live Streamlit demo and open-source code. Future work explores LSTM and attention mechanisms.

• •Built 1D CNN for exoplanet classification from NASA Kepler light curves
• •Achieved 0.96 ROC-AUC by handling class imbalance and data leakage correctly
• •Shared working demo and full source code with GitHub

Generated with AI, which can make mistakes.