Stellar Classification Project

A comprehensive machine learning project for classifying celestial objects (stars, galaxies, and quasars) using photometric data from astronomical surveys.

Overview

This project implements and compares multiple machine learning algorithms to classify stellar objects into three categories:

• GALAXY (0): Extended objects composed of many stars
• QSO (1): Quasi-stellar objects (quasars)
• STAR (2): Point-like stellar objects

The classification is performed using photometric features including magnitude measurements in different filters (u, g, r, i, z) and redshift data.

Dataset

The project uses a local stellar classification dataset (stellar_classification.csv) containing photometric measurements with the following key features:

• u, g, r, i, z: Photometric magnitudes in different filters
• redshift: Cosmological redshift measurement
• class: Target variable (GALAXY, QSO, STAR)

The dataset is included in the repository for easy access and reproducibility.

Data Preprocessing

• Removal of unnecessary attributes (object IDs, coordinates, etc.)
• Handling of extreme outliers in photometric measurements
• Class encoding (GALAXY: 0, QSO: 1, STAR: 2)
• SMOTE resampling to address class imbalance

Machine Learning Models

The project implements and compares five different algorithms:

1. Logistic Regression

   • Hyperparameters: C=10, penalty='l2'
   • Linear classification approach

2. K-Nearest Neighbors (KNN)

   • Optimal neighbors: n_neighbors=1
   • Instance-based learning

3. XGBoost

   • Hyperparameters: learning_rate=0.1, max_depth=5, n_estimators=200
   • Gradient boosting ensemble method

4. Naive Bayes

   • Gaussian Naive Bayes implementation
   • Probabilistic classification

5. Random Forest

   • Hyperparameters: max_depth=None, min_samples_split=2, n_estimators=100
   • Ensemble decision tree method

Key Features

Data Analysis

• ✅ Comprehensive exploratory data analysis (EDA)
• ✅ Statistical measures (mean, median, mode, variance, etc.)
• ✅ Visualization of data distributions and relationships
• ✅ ECDF vs CDF analysis for distribution comparison

Model Optimization

• ✅ GridSearchCV for hyperparameter tuning
• ✅ 5-fold cross-validation
• ✅ SMOTE for handling class imbalance
• ✅ Learning curve analysis

Evaluation Metrics

• ✅ Accuracy and Recall scores
• ✅ ROC-AUC analysis
• ✅ Confusion matrices with heatmap visualization
• ✅ Detailed classification reports
• ✅ Model performance comparison

Project Structure

Stellar-Classification/
├── README.md
├── Stellar_Classification_Report.ipynb    # Main analysis notebook
├── requirements.txt                       # Python dependencies
└── stellar_classification.csv             # Dataset file

Installation

1. Clone the repository:

git clone https://github.com/wyattmog/Stellar-Classification.git
cd Stellar-Classification

2. Install required dependencies:

pip install -r requirements.txt

Usage

Open and run the Jupyter notebook:

jupyter notebook Stellar_Classification_Report.ipynb

The notebook contains:

• Local data loading from CSV file
• Data preprocessing steps
• Exploratory data analysis with visualizations
• Model training and hyperparameter optimization
• Performance evaluation and comparison
• Learning curve analysis

Results

The project provides comprehensive model comparison including:

• Performance metrics for all five algorithms
• Confusion matrices for each model
• ROC curves and AUC scores
• Learning curves showing training vs. validation performance
• Visual comparisons of accuracy and recall scores

Technologies Used

• Python 3.x
• Pandas - Data manipulation and analysis
• NumPy - Numerical computing
• Scikit-learn - Machine learning algorithms and metrics
• XGBoost - Gradient boosting framework
• Seaborn & Matplotlib - Data visualization
• Imbalanced-learn - SMOTE for handling class imbalance
• Statsmodels - Statistical analysis

Future Improvements

• Feature engineering with astronomical domain knowledge
• Deep learning approaches (neural networks)
• Ensemble methods combining multiple models
• Cross-validation with astronomical survey-specific splits
• Integration with additional astronomical catalogs

License

This project is available under the MIT License. See LICENSE file for details.

Acknowledgments

• Astronomical survey data providers
• Scikit-learn and XGBoost communities
• Open source astronomy tools and libraries