Machine Learning techniques to discover and understand the population of flare stars in MeerLICHT data - Data Sci Minor Dissertation

Aug 4, 2026

Upload by Logic

Machine learning methods are applied to discover and understand the population of flare stars using the MeerLICHT dataset. The study introduces the astrophysical background required to interpret stellar radiation, luminosity, flux, and magnitudes, and reviews variable-star categories relevant to the problem. It then details supervised learning approaches (random forests and artificial neural networks) and unsupervised learning via clustering. Model performance is assessed with confusion matrices and ROC curves, followed by feature selection and interpretation of feature importance. Results compare outcomes on PLAsTiCC and MeerLICHT data and outline directions for future work.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages100

Reading Time4 h 12 min

AccessFree to download

Contents

Introduction

Astronomy

Data sources and properties

Machine learning

Results and Discussions

Conclusions and Future work

References

Appendices

University of Cape Town Department of statistical sciences AST5005W Data Sci: Minor Dissertation Machine Learning techniques to discover and understand the population of ﬂare stars in MeerLICHT data Name: Aphiwe Bangiso Student No: BNGAPH002 Supervised by • Prof. Paul Groot Department of Astronomy 18 October 2022 Co-supervised by • Dr. David Buckley Department of Astronomy • Dr. Cole Johnston Radboud University The copyright of this thesis vests in the author. No quotation from it or information derived from it is to be published without full acknowledgement of the source. The thesis is to be used for private study or noncommercial research purposes only. Published by the University of Cape Town (UCT) in terms of the non-exclusive license granted to UCT by the author. Acknowledgments I would like to thank my supervisors for their continuous support during the course of this project, it would not have been possible without their support. I would like to also thank the Inter-university Institute for Data-Intensive Astronomy (IDIA) for providing computational resources for this project. This project was funded by the National Astrophysics and Space Science Programme (NASSP) and the department of astronomy at UCT. Contents 1 Introduction 1 2 Astronomy 4 2.1 Stellar radiation ............................. 5 2.2 Luminosity and ﬂux .......................... 7 2.3 Absolute and apparent magnitude ................... 8 2.4 Hertzsprung-Russell Diagram ..................... 8 2.5 Variable stars .............................. 10 2.5.1 Cepheids ............................. 10 2.5.2 RR Lyrae ............................ 10 2.5.3 Eclipsing binaries ........................ 11 2.5.4 Supernovae ........................... 11 2.5.5 Active galactic nuclei ...................... 11 2.5.6 Tidal disruption events ..................... 12 2.5.7 M type stars ........................... 12 2.5.8 Kilonovas ............................ 12 2.5.9 Mira variables .......................... 12 2.5.10 Microlensing ........................... 12 3 Data sources and properties 13 3.1 PLAsTiCC data ............................. 14 3.1.1 Class distribution ........................ 16 3.2 MeerLICHT sources .......................... 19 3.3 Time series data and feature deﬁnitions ............... 22 3.3.1 Data features .......................... 25 4 Machine learning 27 4.1 Supervised learning techniques ..................... 27 4.1.1 Random forest ......................... 27 4.1.2 Node impurity measures .................... 30 4.1.3 Artiﬁcial neural networks .................... 30 4.1.4 Types of activation functions ................. 32 4.1.5 Optimization algorithms .................... 34 4.2 Model evaluation ............................ 35 4.2.1 Confusion matrix ........................ 36 4.2.2 The receiver operating characteristic curve .......... 38 4.3 Unsupervised learning technique .................... 38 4.3.1 K-means clustering ....................... 39 4.3.2 Hierarchical clustering ..................... 39 5 Results and Discussions 40 5.1 PLAsTiCC ............................... 41 5.1.1 Cross validated results ..................... 45 5.1.2 Learning curves ......................... 50 5.1.3 Modiﬁed PLAsTiCC data ................... 52 5.2 MeerLICHT ............................... 56 5.2.1 The ROC curve ......................... 61 5.2.2 Learning curves ......................... 62 5.2.3 Clustering M dwarf stars .................... 65 5.3 Feature selection ............................ 69 6 Conclusions and Future work 73 7 References 75 8 Appendices 86 8.1 Class distribution ............................ 86 List of Figures 2.1.1 Simulated black body curve ...................... 7 2.4.1 Hertzsprung-Russell Diagram [Agrawal 2018] ............. 9 3.1.1 Samples in each class .......................... 16 3.1.2 Kernel

FAQs

What datasets are used to study flare stars in this dissertation?

The dissertation uses PLAsTiCC data and MeerLICHT sources, including time series data with defined features to support learning and evaluation.

Which machine learning approaches are included?

It covers supervised learning using random forests and artificial neural networks, and unsupervised learning using clustering methods such as k-means and hierarchical clustering.

How is model performance evaluated?

Performance is evaluated using confusion matrices and receiver operating characteristic (ROC) curves, along with learning curves and related diagnostic plots.

Aug 4, 2026

Upload by Logic