Coursework · Shipped · 2025
EPL match predictor
eight ML models compared on one Premier League split
Problem
Given twenty years of Premier League matches, how much of a result can you actually predict, and what moves the needle more: the model you pick, or the data work behind it?
Approach
The team engineered 39 features for goals, streaks, differentials, and form, then trained eight model families on the same train/test split so the comparison stayed fair.
Outcome
XGBoost reached the top reported accuracy on that split, with SVM and logistic regression close behind, and the real lesson was that cleaning, imputation, and feature work moved results more than swapping model families. Coursework context, not a claim of predictive edge.
Screens
