Stellar Classification

Project Overview

This project aims to classify celestial objects from the Sloan Digital Sky Survey (SDSS) into one of three categories: Star, Galaxy, or Quasar (QSO). By analyzing a dataset of photometric measurements (magnitudes) and other astronomical properties, the goal is to develop a machine learning model that can accurately distinguish between these cosmic objects. This is a foundational task in astrophysics for cataloging and studying the universe.

---

Technical Highlights

• Dataset: Kaggle - Stellar Classification Dataset - SDSS17
• Size: 100,000 entries, 18 columns.
• Key Features:
  • Photometric Magnitudes: u, g, r, i, z.
  • Other Properties: redshift, alpha, delta, run_ID, cam_col, field_ID, etc.
• Approach:
  • Data Cleaning: Irrelevant columns that are unique identifiers (obj_ID, alpha, delta, run_ID, rerun_ID, cam_col, field_ID, fiber_ID) were dropped.
  • Outlier Handling: The Local Outlier Factor (LOF) algorithm was used to detect and remove outliers from the dataset.
  • Exploratory Data Analysis: Histograms, box plots, and a pairplot were used to visualize data distributions and relationships between features and the target class. A heatmap was generated to show feature correlations.
  • Handling Class Imbalance: The original dataset is highly imbalanced (Galaxy: 50000, Star: 49978, QSO: 22). SMOTE was applied to balance the classes.
  • Data Scaling: StandardScaler was applied to all feature columns to ensure they are on a similar scale.
  • Multi-class Classification: The target variable class has three distinct categories: 'Star', 'Galaxy', and 'QSO'.
  • Models Used:
    • Logistic Regression, Ridge Classifier, SVC, Random Forest, XGBoost, AdaBoost, Gradient Boosting, Bagging, Decision Tree.
• Best Accuracy:
  • 97.7% with XGBoost Classifier.
  • 97.4% with Random Forest Classifier.
  • 96.3% with AdaBoost, Gradient Boosting, and Bagging classifiers.

---

Purpose and Applications

• Automated Astronomical Object Classification: Enables astronomers to quickly and accurately classify millions of celestial objects from large surveys.
• Catalyst for Discovery: Assists in identifying new quasars, stars, and galaxies, accelerating research in cosmology and stellar evolution.
• Research Tool: Provides a foundational model for studying the properties and distribution of different cosmic objects.
• Data-driven Science: Demonstrates the power of machine learning in processing and making sense of massive astronomical datasets.

---

Installation

Clone the repository and extract the data from the zip file.

Install the necessary libraries:

pip install pandas numpy seaborn matplotlib scikit-learn imbalanced-learn xgboost

---

Collaboration

We welcome contributions to improve the project. You can help by:

• Performing comprehensive hyperparameter tuning and cross-validation for the top-performing models to ensure robustness.
• Exploring the use of different outlier detection and handling methods.
• Investigating the impact of the SMOTE technique on the model's performance on the original, imbalanced data.
• Adding explainability (e.g., SHAP or LIME) to understand which photometric magnitudes are the most critical predictors of each celestial class.