Can Machine Learning Predict the World Cup?

A machine learning model is developed in R to predict football match outcomes using a dataset of 49,000 international matches and team Elo ratings. The project compares the performance of several models, including multinomial regression and LightGBM, to forecast results for events like the World Cup. A significant challenge identified is the models' systematic failure to predict draws, often overconfidently predicting a home win instead. To address this, new features are engineered to capture draw likelihood, such as the absolute difference in team ratings and recent draw frequencies.