Football multi-models

[Data]

I've been experimenting with artificial intelligence engines in an attempt to formulate a profitable strategy in the correct scores markets. Beating the market odds though is never going to be an easy task.

The current stage of this research has reached a point where I'm combining three models to take a consensus of them all to indicate highly rated home wins, draws or aways. One model looks at the home team's record and ranks the chances of H,D or A, a second performs the same task with the away side's record, and a final model that looks at both records to reach a conclusion. Because of this consensus far fewer selections are generated, the majority of which are for Home wins, Away wins to a lesser extent and I've yet to see them all agreeing on a Draw ~ maybe justifying my view that a tied result is the most difficult results market to beat.

However, sticking with my original quest of correct scores, as well as a straight H, D or A bet, each selection is also bet as a correct score using the league's most common scoreline.

I'll use the same level stake for each wager, because HDA has a high strike with poor odds (most will be odds-on, but I'm looking for 80% strike) whereas correct scores have better odds but a lower strike rate. My aim is a minimum 10% yield for each bet type. But, the best laid plans . . .

Bets for this weekend are;

ENG Middlesbro v Swansea    1-0
ENG Nottm Forest v Rotherham    1-0
BRA Flamengo v Ath Mineiro    1-0
CHI Dalian Aerbing v Shanghai Shenhua    2-1

10 pt home win on each.
10 pt correct score on each.
 

[liero1]

interesting, following this!

could you show the predicted probability of your bet selection if possible?

cheers!