Library
ep
School · 2025
EPL match predictor
eight models against twenty years of Premier League data
8 models, one split
39 engineered features
~99% top accuracy (xgboost)
20 yrs of match data
model accuracy comparison 
feature correlation heatmap 
xgboost results About
A group project predicting English Premier League outcomes from 20+ years of match data, using 39 features covering goals, streaks, differentials, and form. Eight models trained on the same split: Random Forest, MLP, Decision Tree, KNN, Naive Bayes, Logistic Regression, XGBoost, SVM.
XGBoost finished at ~99% accuracy with SVM and Logistic Regression close behind. Feature engineering moved accuracy more than model choice.
Liner notes
- Same train/test split across all eight models for a fair comparison.
- Cleaning, imputation, and feature work drove most of the gains.
- Groundwork for the later sports-ML and quant projects.
Credits
Language python
Models 8-way comparison
Winner xgboost ~99%
Data kaggle · 20 yrs epl