Group: Juri Soiri, Tair Kuanysh

Aim: To test and apply machine learning methods to predict goal outcomes in football matches. We specifically compared Polynomial Regression against a Multi-Layer Perceptron (MLP) to see which could better manage the unpredictable nature of sports data.

What we did: We utilized Pandas, Scikit-Learn and Numpy to process five seasons worth of league data from across Europe. We built the second-degree polynomial regressor and engineered the MLP regressor with four hidden layers (15 neurons) using the ReLU activation function.

What we learnt: While the MLP is a more sophisticated model, Polynomial Regression provided superior interpretability. The MLP model achieved a lower Mean Squared Error (MSE), suggesting a higher capacity for capturing complex patterns, though the results demonstrated a risk of overfitting due to dataset size. The project illustrated the trade-off between model complexity and generalizability.