Aim is to discover the application of neural networks in AFL games and explore sets of features that can be used for predicting the current game

Two datasets were created to explore which features are useful for predicting match outcomes using two separate neural networks. The first model was tested on a Rating Systems dataset (11 features) which is an Elo model with Home Advantage. The Elo dataset is based on an optimisation created in Excel solver using a regression towards the mean for each new season to reset the Elo.

The second model was tested on a Performance based dataset which contains the average performance in a previous match (21 features). The neural network was created with Python using Keras . Grid Search cross fold validation used to find optimal hyper-parameters for both models and a Drop Out layer added to eliminate overfitting