AstroClassifierML is a machine learning project aimed at classifying space objects. It uses data from Sloan Digital Sky Survey, applying advanced machine learning techniques to classify Galaxies, Stars, and Quasars.
- TensorFlow and Scikit-Learn for Machine Learning: This project employs TensorFlow for building and training neural network models and Scikit-Learn for SVM model implementation and data preprocessing.
- Pandas for Data Handling: Utilises Pandas for data manipulation, enabling efficient data operations.
- Data Preprocessing: Features data preprocessing techniques like feature scaling, outlier removal, and handling imbalanced datasets using SMOTE.
- Model Optimisation and Evaluation: Uses the Adam optimiser for the neural network and evaluates models based on accuracy metrics.
- Regularisation Techniques: Applies dropout in neural networks to prevent overfitting and ensure model generalisability.
- Deep Neural Network (DNN) Model:
- Architected with dense layers and dropout for regularisation.
- Utilises Early Stopping to monitor training performance.
- Support Vector Machine (SVM) Model:
- Implemented using Scikit-Learn's
svm.SVCwith an RBF kernel. - Optimised with hyperparameter tuning.
- Implemented using Scikit-Learn's
main.py: Main script for executing the machine learning workflow, including data preprocessing, model training, and evaluation.data_handler.py: Script for data extraction and preprocessing. Handles tasks like data cleaning and feature engineering.models.py: Contains the implementation of the DNN and SVM models.plotter.py: Handles all the plots generated from the models and data.
The data contains a range of characteristcs of Stars, Galaxies and Quasars, which can be visualised here:
The script outputs the accuracy of the DNN model and generates a plot showing the training and validation accuracy of the model over epochs.
Using the SVM model we can categories the errors in the model seen here:
Sloan Digital Sky Survey