Skip to content

README.md

Latest commit

 

History

History
145 lines (95 loc) · 4.63 KB

README.md

File metadata and controls

145 lines (95 loc) · 4.63 KB

🌌 Star Classifier Model using TensorFlow and K-Means Clustering

Overview

This repository presents a scientific machine learning pipeline for stellar classification and exploration of stellar populations, inspired by astrophysical data analysis practices commonly used in space science institutions.

The project combines:

  • Supervised Deep Learning (TensorFlow / Keras) for classifying stars into physical stellar classes.
  • Unsupervised Learning (K-Means Clustering) to visualize stellar populations on the Hertzsprung–Russell (HR) Diagram, revealing natural groupings driven by stellar evolution.

The model classifies stars based on physical parameters such as surface temperature, relative luminosity, relative radius, color, and absolute magnitude, and demonstrates how machine learning can faithfully recover known astrophysical structures.

Repository Structure

STAR CLASSIFIER/
├── kMeans_model.sav           # K-Means Clustering Model
├── Stars.csv                  # Stellar dataset (from Kaggle)
├── train_model.py             # Model training script
├── evaluate_model.py          # Model evaluation and metrics generation
├── HRdiagram.py               # HR diagram generation and clustering
├── data_info.py               # Dataset inspection and metadata
├── scaler.pkl                 # Saved feature scaler
├── star_classifier_model.h5   # Trained TensorFlow model
├── confusion_matrix.png       # Confusion matrix visualization
├── roc_curve.png              # ROC curves and AUC
├── HR diagram.png             # HR diagram with K-Means centroids
├── dict.json                  # Label mapping and metadata

Model Evaluation Results

The trained neural network achieves very high classification performance on the stellar dataset, with:

  • Near-perfect accuracy on the validation set
  • Stable convergence after 150 epochs
  • Low categorical loss values

Evaluation metrics include:

  • Accuracy
  • Precision
  • Recall
  • F1-score
  • Confusion Matrix
  • ROC Curves and AUC (multi-class)

These results indicate that stellar classes are strongly separable in the chosen feature space, consistent with astrophysical expectations.

HR Diagram and K-Means Clustering

The repository includes an unsupervised clustering analysis using K-Means (k = 6) to generate an HR diagram.

Axes used:

  • X-axis: log10(Temperature)
  • Y-axis: Absolute Magnitude

K-Means clustering automatically identifies stellar populations corresponding to:

  • Brown Dwarfs
  • Red Dwarfs
  • White Dwarfs
  • Main Sequence Stars
  • Supergiants
  • Hypergiants

The resulting Silhouette Score ≈ 0.63, indicating strong and physically meaningful clustering. Cluster centroids are visualized directly on the HR diagram.

HR_Diagram

Dataset

The dataset used in this project is publicly available on Kaggle: "https://www.kaggle.com/datasets/brsdincer/star-type-classification"

The dataset contains labeled stellar data derived from astrophysical catalogs and is widely used for educational and research purposes.

Model Architecture

The neural network architecture is intentionally simple and interpretable:

  • Input Layer: Stellar features
  • Hidden Layer 1: Dense(64), tanh
  • Hidden Layer 2: Dense(32), tanh
  • Hidden Layer 3: Dense(16), Softplus
  • Output Layer: Dense(6), Softmax

Training Configuration:

  • Optimizer: Adam
  • Loss Function: Sparse Categorical Crossentropy
  • Metrics: Accuracy
  • Epochs: 150
  • Batch Size: 24

This architecture balances expressive power with computational efficiency and aligns well with the physical structure of the data.

Model Performance Summary

  • Classification Accuracy: ~100%
  • Stable loss convergence
  • High separability across stellar classes
  • Strong agreement between supervised labels and unsupervised clustering

These results demonstrate that machine learning can successfully capture the underlying physics of stellar evolution.

Applications

  • Stellar classification pipelines
  • Educational demonstrations of HR diagrams
  • Astrophysical data analysis
  • Feature engineering experiments for scientific ML
  • Portfolio project for scientific machine learning

Author

Developed by Jovan, an independent researcher with a background in physics and scientific computing, exploring the intersection of astrophysics and machine learning.

License

This project is released under the MIT License, allowing free use, modification, and distribution for academic and commercial purposes.

"Machine learning does not replace physics — it reveals it." 🌠