As some of you have noticed, for every game this season, I have posted a set of 3 predictions ( I actually calculated 5) just to see how 2 general AI apps, would do at predicting our results. As a basis to compare those predictions I also decided to use 1 general random prediction method, along with 2 fairly random methods, just to see how well or poorly Ai predictions could do. Method 1 - required me to make some choice. Every game would be 1-0, there would be no draws. I just needed to decide if we would win or whether we would lose. Method 2 - was based upon last seasons results in game order. So whatever the result of our first home game last season, would be the prediction of our first home game this season, where as the result of our fourth away game last season, would be the prediction for our fourth away game this season. ( I hope this fails for the game against Plymouth as it would mean a 4-0 defeat for ourselves ) Method 3 - This was the one I used when posting my predictions in the match day thread, and was based upon whatever score we had last season against the opponents was the score that I predicted this seasons. For those teams that got promoted and relegated last season, they were replaced by whichever teams got relegated from the EPL and promoted from league 1. Thus Leicester, Ipswich and Southampton were replaced by Luton, Burnley and Sheff Utd, whilst Portsmouth, Derby and Oxford, replaced Birmingham, Huddersfield and Rotherham. The important thing for my 3 predictions methods is that it had to ignore form, league position, signings, injuries etc, and most importantly, I could do the predictions for all our games, prior to the start of the season. The two AI's chosen at the start of the season were ChatGPT, probably the most well known of the general AI apps, and Google Gemini. Gemini was chosen as out of all the other AI's I tried this was the only other one apart from ChatGPT, that were willing to provide a prediction. The other sites were not prepared to support what they considered could be used for betting. After week 30 Gemini, decided that it no longer was able to do predictions, but luckily I found an alternative in Ninja AI that would work. Predicting Correct Scores Both ChatGPT and method 2 of predicting scores by last seasons game week, managed to get 6 results perfectly correct out of 45 games. Method 3 of predicting scores by last season opponents managed to get 5 correct scores. Both Gemini/Ninja and method 1 (1-0 wins or losses) did the worst with just 3 correct scores. Correct Match Outcome Gemini/Ninja got the most correct match outcomes with 25, with ChatGPT and my 1-0 method having 22 correct outcomes, whilst methods 2 and 3 got 18 and 15 match outcomes correct. Overall Outcome Both ChatGPT and Gemini/Ninja got 17 matches incorrect. However given that ChatGPT got double the correct scores of Gemini/Ninja, then I would have to declare ChatGPT the winner. My 3 methods got 20, 21 and 25 results incorrect. However whilst all my 3 methods got more match outcomes incorrect, 2 of those methods got more correct scores than Gemini, so it becomes difficult to compare, unless you give an arbitrary value to a correct score Summary AI performed slightly better than the 3 random methods I selected, but still has a long way to go to improve. Obviously predicting football scores is not a science, as there is too much unpredictability in any one game, but I would imagine that at present anyone who actually took the time to study form etc would do better than AI has done this season. and of course for anyone who wants to see my data then its enclosed below.
