After determing some disadvantages of the Poisson model, I started using Tensorflow to create my first own model based on neural networks. I called it the Team Strength MLP, as the base variables are the same as for the Poisson model. These variables correlate with the relative strength of a team to the league average. The first version of the model just uses the ability of a team to score and prevent goals. Later on, this can be extended by several variables for more detailed strength information.
In following blog posts, I analyse the picks selected by the Team Strength MLP model to get more insights about the weaknesses of my model:
- Bundesliga match days 1 – 10 (2018/19)
- Four things I have learned after using a neural network for 6 months
- Retrospective for Bundesliga season 2018/19
Pick history stats for the Bundesliga season 2018/19.
If you have further questions, feel free to leave a comment or contact me @Mo_Nbg.