Colossl Victory

So now using version 3 a summary of its performances over recent weeks and its predictions for the upcoming games Week 17-23  Review - Week 17 - 7 results with 2 scores Week 18 -  3 results with 0 scores Week 19 -  6 results with 1 score Week 20 -  5 results  with 2 scores Week 21 -  3 results with 0 scores Week 22 -  6 results with 2 score s Week 23 -  7 results 0 scores Overall a fairly good set of results with just a few poor weeks in the middle. Scores very low with this model but getting a clear improvement in results. Running just above the average for last season with 37/70 results. Week 23 Predictions - Home Predicted Score Away Arsenal 2-0 Cardiff Fulham 1-0 Brighton Huddersfield 1-0 Everton Wolves 2-1 West Ham Man United 1-2 Burnley Newcastle 0-1 Man City Bournemouth 1-0 Chelsea Southampton 1-0 Crystal Palace Liverpool 2-0 Leicester Tottenham 2-1 Watford A huge surprise there with Burnley to win away at Man ...

Version 3 The new predictive element I have calculated to be the best to use is a logistic regression machine learning element to predict result of a match. Followed  by taking the average xG scored and conceded by both the home and away team to produce an average value for the home team to score and away team to score which then uses a Poisson distribution with these numbers as mean to find the most likely score line with matching result to the first element. Naive Bayes and Results was selected by testing performance against 3 other predictors that use either Bayes or Logistic regression to predict either scoreline or result. The performance test was accuracy of predicting the correct result even if scoreline was used to train the data as ultimately that is what they are both trying to predict. Results shown below: ML\Predict Results Score Bayes 0.47 0.31 Logistic 0.50 0.47 So this is only a small improvement from the old version 2 predictor which got 0.46...