Nice thought! Just as you said, because the equation is very simple we could aprroach the problem with linear regression. However, the main idea of the article is not only to find the best solusion but to show the possibility of representing learning model to real physical equation. In discrete choice analysis, by arranging neural network we can represent utility model. Plus, with X-AI technique, such as SHAP, we can interpret the neural-network better. Not sure if we can do that with linear regression..