AP Research: A Solubility Prediction Algorithm for Salt in Water at Different Temperatures
Advisor: Joselyne Campos
This study aims to conduct empirical research involving human subjects to evaluate the efficacy of my algorithm for solubility prediction in comparison to established methodologies in the field and a control group that receives no solubility prediction assistance. The objective is to assess the relative utility of my algorithm against more traditional scientific approaches, thereby contributing to the understanding of its practical applicability in solubility prediction.
Project Blog
Week 0
Hello,
My name is Marcus Persson, and I am doing a senior project on how people could benefit from culinary science-focused software. I am starting small by building a machine learning model that models solubility for salt in water at different temperatures (this is useful for some recipes and salt crystals, but this is really just a starting point for future inquiry. I will then gather a focus group of students and compare my user-experience-focused algorithm to other solubility approaches and a group without assistance. This is challenging for me because I have never conducted human subject research.
I was inspired to do this because I noticed that many cooking recipes left no room for creativity. I think that software like this can allow people to experiment with what they cook while still having the reassurance that their meals will turn out well.
Thank you!
Weeks 1-4: Project Pivot
Hello again,
My project shifted slightly over the past few weeks—positively. At first, when I showcased my senior project, I planned on implementing a machine learning algorithm that predicted water's solubility in salt at different temperatures and comparing it afterward through user testing within a focus group. The idea was to determine if a more user-experience-friendly algorithm would aid users with their creative cooking challenges.
But the more I ventured into the technical and logistical ends, the more I realized that doing human subject research would be more complex than expected. From IRB clearance to creating an effective study and working around ethical requirements, I found that much of my time would be robbed from what I most eagerly anticipated: building the algorithm itself.
So I changed my direction. I chose to put my energy into tuning the machine learning algorithm, and I'm happy to report that I've finished building it!
The model uses a simple, data-driven approach to predict the solubility of table salt (NaCl) in water at various temperatures. I taught it using publicly sourced solubility data and set out to build something lightweight, open, and accessible—even to the non-chemist. Results? It's not the best, but it's reasonably accurate, and that's a victory considering the narrow scope and tiny dataset I was working with. The algorithm can now return reasonably close approximations and may form the basis of more sophisticated, general-purpose kitchen science tools.
With the model now complete, I've been dedicating my efforts to crafting the research paper that will outline the process from the issue I was attempting to resolve to how the algorithm was trained and tested. In addition, I will include a discussion on the transition to human testing, what was gained from this shift, and how it altered the course of my project.
In the end, though, the emphasis has changed while the project's purpose has remained the same: developing tools that will better make people creative and more self-assured in the kitchen by using science in a useful and accessible manner.
Thanks for hanging in there with me—I'll be glad to post more when I finish the paper!
Post 3: Last Week
Blog Post: What I Learned from My AP Research Project
For my AP Research project, I studied how machine learning can be used to predict solubility in mixtures of salts and solvents -an introduction to computational chemistry for me. I wanted to make chemistry more accessible using simple models and free data.
I built and tested different machine learning models using data sourced from various scientific and industrial sources. My best model was pretty accurate, and it showed that even with limited data, it’s possible to make accurate predictions without an ideal solution assumption.
What I Learned
This project taught me a lot. I got better at finding data, writing scientific papers, and conducting a literature review. It also showed me how important it is to curate data. Things didn’t go as planned a few times — the data I needed wasn’t available, or I had to change how I built the model. This showed me to think of my feet.
The Presentation
The presentation went well. I explained my topic clearly and answered questions with confidence. But if I could change one thing, I would’ve used more images and visual examples. That would’ve helped people understand my points more easily.
What I Would’ve Done Differently
-I should’ve started collecting data earlier. Cleaning and organizing it took more time than I expected.
-I wish I’d added more images in the presentation to explain key ideas better.
-I also could’ve asked for feedback from someone with experience in computational chemistry once I finished the model and wrote the first drafts of the paper.
Future Plans
In college, I want to continue exploring computational chemistry, but for different purposes, like designing new materials or studying reactions that are hard to test in the lab. This project was a great first step.