Exoplanet Habitability Detector

The pace of exoplanet discovery now exceeds the capacity of manual screening, making early habitability assessment difficult. We present a data-driven system that combines a preprocessing model with a neural network trained on 5032 confirmed exoplanets from the PHL Arecibo Catalog and NASA Exoplanet Archive. The model condenses more than 30 astrophysical variables, such as stellar flux and density into a compact numerical representation. The classifier assigns each planet probabilities across 3 habitability categories: Not Habitable, Somewhat Habitable, or Very Habitable. On the independent test set (n = 613), the model achieved a macro F1 score of 0.517 and an overall accuracy of 91%. This reduces screening time from hours to seconds. A lightweight API links the trained model, preprocessing pipeline, and features, with predictions returned instantly via CSV upload or form entry. A React/WebGL front-end renders interactive 3-D models, heat maps, and probability charts.