Differences in Men's and Women's March Madness Machine Learning Prediction Features

Last year, I entered the Kaggle March Madness machine learning competition for the first time. It threw me for a bit of a loop that a single competition has you submit predictions for both the men's and women's tournaments. It didn't seem safe to assume that the same features would predict both men's and women's games optimally, but due to time constraints, I just trained a model on men's data so I could also use it to fill out work and friends' bracket competitions, then reused the exact same model for women's predictions in the Kaggle competition.

I've started adjusting the model for 2026, and decided to test my intuition that men's and women's predictions would benefit from tailored models. I took the same feature set, trained two logistic regression model instances separately on men's and women's data, and compared the weights.

Here are the highlights:

1. Women's games seem to be more predictable: the women's model test results were about 10% more accurate (~70% accuracy for men and ~80% for women)
2. The women's tournament seems to have fewer upsets: seeds are by far the most important feature for women
3. My models seem to dislike the three-point revolution, especially for women: shooting threes seems to be a neutral weight for men, but heavily negative for women — and getting to the free throw line is a negative factor for both