Evaluating the Performance of Machine Learning Models in Remote Sensing for Sustainable Development Goals: A Meta-Analysis
Foreword
The aim of this research was to assess whether study features can explain variations in results across studies. To my knowledge, this is the first study to apply weighted meta-analysis techniques to examining the performance of machine learning models in remote sensing. While I intended to adhere to PRISMA guidelines, the exploratory nature of the topic—and my own learning journey—meant pre-registration was ultimately not conducted, and with data extraction carried out solely by myself, there is a degree of subjectivity and potential for error there. But I’m giving away spoilers for the discussion, so I’ll stop myself there!
I have succeeded in building this manuscript using Quarto, with minimal stylistic adjustments after rendering. The entire code for this project is available on GitHub, and an HTML version with integrated code chunks can be accessed on GitHub Pages website. The data processing and paper selection analysis scripts are all available on the GitHub-hosted site under the appendix (more details about the file organisation are available on the GitHub page). I have also integrated the Leiden University master thesis cover format into the Quarto book, so if any future student would like to reuse it please do! I toyed with the idea of creating a Quarto book template but that is a project for another day.
One last thing, in the discussion of this thesis I suggest that journals should begin requesting data submissions alongside the manuscript to enable active, ongoing meta-analyses. In support of this idea, I developed a small pilot website to demonstrate how such a system might work. If anyone feels inclined to add to the dataset, there are instructions on how to do that there.
To the reader: thank you for taking the time to read my thesis— there’s still time to stop reading, and I won’t know any different! If you’ve made it this far —hi supervisors, and independent reader (, and mum!?)— I would like to apologize in advance for continuing the convention of inconsistent notation across meta-analysis research. I can only hope I have been consistent within my own work as its best not to change notations \(\mu\)-dstream.