Stellar Classification Project

Overview

This project aims to classify astronomical objects into three categories: Stars, Galaxies, and Quasars (QSOs) using data from the Sloan Digital Sky Survey (SDSS). The dataset contains 100,000 observations, each described by 17 features and a class label. Machine Learning (ML) and Deep Learning (DL) models are implemented to achieve accurate classification.

Dataset

Stellar Classification Dataset - SDSS17

• Source: Kaggle
• Features: 17 features including magnitudes and colors, with a target class (Star, Galaxy, QSO).
• Size: 100,000 observations.

Citation

fedesoriano. (January 2022). Stellar Classification Dataset - SDSS17. Retrieved [Date Retrieved] from Kaggle.

Acknowledgements

The data released by the SDSS is under public domain. It is taken from the current data release RD17. More information about the license: SDSS License Information

Methodology

Preprocessing

1. Feature Engineering: Added derived features like "G-I" (difference between g and i magnitudes) and "Z-G" (difference between z and g magnitudes).
2. Feature Scaling: Standardized all feature values using StandardScaler.
3. Class Balancing: Used class weights to address imbalanced data.

Models

1. Multi-Layer Perceptron (MLP): Fully connected neural network with dropout layers for regularization.
2. Convolutional Neural Network (CNN): 1D CNN with convolution and max-pooling layers, designed for sequential data.
3. Long Short-Term Memory (LSTM): LSTM network to capture sequential dependencies in features.

Training and Evaluation

• Train-Test Split: 80% training, 20% testing.
• Validation Split: 5% of the training data used for validation.
• Metrics: Accuracy, confusion matrix.
• Optimization: Categorical cross-entropy loss with the Adam optimizer.

Results

Each model’s performance was evaluated, and the best-performing model was selected based on accuracy. Confusion matrices were used to analyze the classification results further.

How to Use

1. Clone the repository.
2. Install required Python libraries: scikit-learn, tensorflow, numpy, matplotlib.
3. Load the dataset and follow the preprocessing steps in the code.
4. Train models and evaluate results.
5. Use the best-performing model for predictions on new data.

License

This project uses publicly available SDSS data, which is under the public domain. For more details, refer to the SDSS License Information.