The British Premier league has 20 teams, they play 380 games in a season, each team plays with each one game at home and one game away. 17 out of these 20 teams were playing in the previous season and all results are available. 3 teams are new, we ignore them and only predict for those that played previous season.
How to predict outcomes in a new season for these 17 teams using historic records and newly obtained results. For example, when several games already played in a new seasons, they can be taken on account. We need to predict the exact score, not only a winner. Winners are obvious in most games. When predicting scores we need predict them as vectors, so if it is 3:2 that does not mean we can predict 3 by one model and 2 by another, it is clear that both numbers are related. Also the model must be probabilistic. So we have to predict probabilities of multiple possible outcomes.
I'm not trying to get quick rich. Bookmakers disallow to use AI and disqualify those who, may use them. Also, it is already known that AI predicts better than humans. Also, bookmakers are simply exclude those who win too much without any explanation, so it is not a money making method, but a research.
The prediction must use only the scores from the previous seasons and not any other insider information, such as traumas, new players, new trainers, weather, rain, wind and so on. These factors are considered as random noise and expressed indirectly in probabilities.
Andrew Polar To predict the outcomes of the new season for the 17 teams using historic records and newly obtained results, we can utilize a probabilistic model that captures the relationships between scores of both teams in a match. A common and effective approach for this type of problem is using a Bayesian hierarchical model. This approach allows us to model the dependencies between the scores of both teams and update our predictions as new data comes in.
I saw this concept, but each time I try it, I face problem. I don't have critical parameters to build the model. Each valid example does have it, but I don't have. So it looks like it is fake or detached from reality and people who promote it, found some non-common examples.
The data you have looks like follows:
Team A at home, team B away 3:2
Team A at home, team C away 1:2
---- and so on
Team K at home, team B away 2:2
Team K at home, team C away 3:1
---- and so on
The answer we want:
Team A at home, team K away
2:2 with probability 0.3
1:2 with probability 0.2
2:1 with probability 0.2
Bayesian hierarchical model needs probabilities all the way down the tree. What are they? We don't have them.
- Badges
- Science topic
- Similar topics
- Indexes