Physics of Data Masters Degree - 1st Semester - Laboratory of Computational Physics - Final Project
This project from the Laboratory of Computational Physics (2022-2023), focuses on the astrophysical phenomenon of binary black holes and their formation. Binary black holes (BBH) are systems of two black holes in orbit around one another, and their mergers are detectable through gravitational waves. The goal of the project is to identify the key features that influence the evolution of binary star systems into binary black holes. By employing machine learning techniques, the project explores various features and their impact on this evolutionary process using a dataset of simulated binary black hole formations.
The project involved analyzing features of binary star systems and applying machine learning algorithms to predict the evolution of these systems into binary black holes. The dataset, which was imbalanced, was preprocessed by balancing paths of evolution and removing outliers. Key features, such as the zero-age main sequence (ZAMS) masses, binary black hole masses, orbital eccentricity, and semi-major axis, were evaluated for their importance.
Three machine learning models were applied: Linear SVM, Random Forest, and Neural Networks. The Random Forest model performed the best, aligning closely with expectations. Feature analysis revealed that masses (both alone and in ratios) and the semi-major axis were the most critical factors in predicting binary black hole formation. The project demonstrates that machine learning can effectively identify the key astrophysical factors that drive binary star systems' evolution into binary black holes.