WhereItLands
End-to-End ML Football Match Predictor & MLOps API
- Engineered a modular ML pipeline adopting OOP and SOLID principles, achieving a 64.5% 3-way match accuracy for predicting World Cup results.
- Developed a StackingRegressor ensemble and bivariate Poisson regression with Coles-Dixon correction for exact score probabilities.
- Processed 14,000+ historical matches to build an OO-ELO rating system calculating team momentum via tournament weights and home advantage.
- Built an asynchronous FastAPI/Uvicorn REST API utilizing a Feature Store pattern with Joblib and Parquet serialization to minimize model loading latencies.
PIPELINE OVERVIEW
v2.4 [PROD]
Dataset Ingestion
14,000+ Matches
Feature Store
Parquet + Joblib
Stacking Ensemble
64.5% Accuracy
API Delivery
FastAPI Async