Exoplanet Detection using Machine Learning

This repository contains the full pipeline for machine learning-based exoplanet detection, specifically for F type stars. The model is designed to predict the existence of exoplanets based on astrophysical parameters derived from NASA's Exoplanet Archive.

Project Structure

├── create_train_file_improved_latest.ipynb        # Processes training data
├── create_test_file_improved_latest.ipynb         # Processes test data
├── exoplanet_prediction_improved_original.ipynb     # Machine learning model for prediction
├── final_train_file.csv                     # Preprocessed training data
├── final_test_file.csv                      # Preprocessed test data
├── exoplanet_archive_test_data.csv         # Raw exoplanet test data
├── exoplanet_archive_test_data_processed.csv # Processed test data
├── labelled_training_star_data_943_samples.csv         # Raw labelled training data
├── labelled_training_star_data_943_samples_processed.csv # Processed training data
├── constraints.txt                          # scipy version constraint

Requirements

Make sure you have the necessary dependencies installed. You can install them using:

pip install lightkurve
pip install -r constraints.txt

Package Versions

This project was built using the following package versions:

- Python 3.x
- TensorFlow 2.x
- Keras 2.x
- NumPy 1.x
- Pandas 1.x
- Matplotlib 3.x
- Imbalanced-learn 0.x
- SciPy 1.14.0
- Lightkurve 2.5.0
- Tsfresh 0.20.3

How to Run the Project

1. Prepare the Data

• Run create_train_file_improved_latest.ipynb to generate the training file with 'labelled_training_star_data_943_samples_processed.csv' as input.
• Run create_test_file_improved.ipynb to generate the testing file with 'exoplanet_archive_test_data_processed.csv' as input.

2. Train & Evaluate the Model

• Run exoplanet_prediction_improved_original.ipynb to train and test the neural network model.
• The script will output various model performance metrics and generate predictions.

3. Predictions

• The model will save predictions as predictions.csv, containing:
  • Star_ID (if applicable)
  • Prediction (1 = Exoplanet, 0 = Not an Exoplanet)

Model Overview

The model used in this project is a deep learning based model built using TensorFlow/Keras. It includes:

• Fully connected layers with ReLU activations
• Binary classification with Sigmoid activation
• Optimized using AdamW optimizer
• Balanced dataset using SMOTE

The model has been trained on F, G and K-type star data.