An examination of Machine Learning Models applied to various fields in Astrophysics

The recent advancements in the field of machine learning has spawned many models of varying complexity and design. In this thesis I showcase and discuss the novel application of various machine learning models to address potentially underperforming methods of observational measurement in the field of astrophysics. The work presented here includes the functionality of a noise-reduction autoencoder to transient images of galaxies, as well as utilising decision trees and random forests to match optical and radio targets across different catalogues. In those works I found that I was able to reduce noise on transient images as effectively as prior methods in a fraction of the time, as well as only using a single image to train our model. I was also able to match optical to radio targets with an accuracy range between 66% - 68% and 68% - 70%, dependent on the models used as well as their settings and features they were trained on. I expand upon my previous works by discussing the history of the observational astrophysics relevant to my works, the models I used for said works, outline in more detail their structure and design, and discuss the metrics I was unable to incorporate due to time constraints. Finally, I give an overview of the background, data, methodology and results of applying machine learning models to predict the metallicity abundance in galaxies whilst investigating the validity of metallicity relations involving various features.