Research / Applied ML
Match Balance Lab
A controlled experiment comparing formula-based and ML approaches to assigning players to teams.
01Players→02Features→03Model→04Teams
Context
Teams that look equal on average can still differ in role and skill composition. The question was whether a trained model could produce more stable assignments than a fixed formula.
How it works
Created a synthetic match dataset, player features and one comparison loop. The CatBoost model and formula-based algorithm share a FastAPI interface, so their outputs can be evaluated under the same conditions.
What it enables
The experiment separates model value from implementation differences: identical inputs, one API and comparable outputs make it possible to see where ML genuinely adds value.