A series of supervised machine learning algorithms to attempt to predict premiere league game results.
Includes two neural networks all tested with hidden sigmoid, tanh and softmax layers and one SVM to classify teams and perform regression on game data to predict results.
Overfitting of the model has caused the majority of issues in the prediction and learning models, a smaller training set and larger testing set was used to avoid overfitting, which has recently led to inaccurate predictions.
Current results from a feed-forward 4-55-1 neural network, with a hidden sigmoid layer trained using back-propagation on a dataset of 152 Tottenham home games (from 2010-2016), with 4 inputs: [homeTeamDefence, homeTeamAttack, awayTeamDefence, awayTeamAttack] -> [result] where all inputs and outputs are a member of [0,1)
[The neural network was trained until the error was ~5% to avoid overfitting]
- Tottenham vs Man United home win probability: 0.53
- Tottenham vs Arsenal home win probability: 0.46
- Tottenham vs Man City home win probability: 0.83
- Tottenham vs Southampton home win probability: 0.74
- Tottenham vs West Brom home win probability: 0.71
- Tottenham vs Liverpool home win probability: 0.63
- Tottenham vs Watford home win probability: 0.77
- Tottenham vs Crystal Palace home win probability: 0.78
- Tottenham vs West Ham home win probability: 0.95
- Tottenham vs Swansea home win probability: 0.81
- Tottenham vs Chelsea home win probability: 0.53
- Tottenham vs Everton home win probability: 0.95
- Tottenham vs Stoke home win probability: 0.83
- Tottenham vs Sunderland home win probability: 0.94
- Tottenham vs Newcastle home win probability: 0.93
- Tottenham vs Norwich City home win probability: 0.91
- Tottenham vs Aston Villa home win probability: 0.87
This is still a proof of concept and very much a work in progress. It is clear that the success of such a neural network is still unproven and predictions may well be arbitrary due to the stochastic nature of the training data.
Work by Thom. 17/06/2016 @thomx128