Beyond the Code: Hitting 79% Accuracy and The Art of Research Documentation

Welcome to another weekly update from my internship at the KMUTT Geospatial Engineering and Innovation Center (KGEO). This past week (September 21st to 25th) was all about refinement by taking the raw models I built previously, pushing their limits, and learning how to translate complex code into a comprehensive research paper.

Presentations and Pushing the Limits

The week kicked off with a progress presentation to Ajarn Parivate and the rest of the GIS class. I showcased my recent work on Above Ground Biomass (AGB) estimation modeling. The session went incredibly well, and the constructive feedback I received gave me a clear direction for future improvements.

Taking those suggestions back to my workstation, I dove into some academic papers and expanded my model's hyper parameter tuning. The effort paid off beautifully: the model's accuracy improved to 79%. Given the current constraints of our data, this is a very solid and acceptable benchmark for remote sensing analysis.

The Importance of Visualizing Mistakes

With a strong model finally in hand, my focus shifted from coding to documentation. I began drafting a full research paper detailing my methodology and findings for Pi Boy. This pre-research analysis is crucial, as it will serve as the foundational assessment for the main project we are rolling out next month.

During the review process, Pi Boy shared a highly valuable piece of advice regarding research writing: a good paper doesn't just highlight the successes. It must clearly and visually explain everything. Including the mistakes and anomalies. Transparently documenting what went wrong provides a much clearer vision and makes the overall explanation of the research significantly stronger.

(On a side note, amidst the heavy research writing, I also had some great discussions with Mr. Agus to plan and strategize for our upcoming website project!)

Broadening the Scope

By Friday, I realized that simply presenting my single "best-performing" algorithm didn't tell the whole story. To make the pre-research analysis truly comprehensive, I went back to the drawing board to assess, display, and perform a deep dive into every single algorithm I tested.

Naturally, evaluating each algorithm individually is a very time-consuming process. However, proving why certain algorithms failed while others succeeded is what makes the research rigorous.

Final Thoughts

This week was a powerful lesson in balancing technical execution with the meticulous art of research writing. I am learning that building a great machine learning model is only half the battle. But being able to explain it clearly and honestly is what truly matters. I am wrapping up the week feeling highly prepared for the real project next month. See you in the next update!