Stellar Object Classification Project

Overview

This project classifies stellar objects into three categories (Galaxy, Star, Quasar) using sensor data. I compared the performance of three models: LightGBM, Random Forest, and XGBoost.

The Data

• Source: https://www.kaggle.com/datasets/fedesoriano/stellar-classification-dataset-sdss17
• Size: 100,000 rows, 17 features (only used 8 features in training).
• Preprocessing: Standard Scaling applied; target labels encoded as integers.

Results & Visualizations

1. Model Comparison

Here is the performance comparison across Accuracy, F1, and Precision.

2. PairPlot

The pairplot for the data.

Key Findings

• LightGBM achieved the best balance of speed and accuracy.
• Random Forest had slightly higher recall and accuracy but was significantly slower to train.
• The "Quasar" class was the hardest to predict due to class imbalance.

How to Run

1. Clone the repo.
2. Install requirements.
3. Run the notebook: stellar_classification.ipynb