A different approach?

[Data]

Many football rating methods are compiled on a match-by-match basis in real time, inasmuch a ratings is assessed for a game at hand, then using the result of said game, the rating is adjusted accordingly. The most recent match therefore has perhaps too much of an influence on how the new rating moves.

This approach presented here also uses past results to calculate the outcome, but it looks at ALL previous game results and treats them as equally important. Trends therefore are not picked up, whichever way a team's form is heading is totally ignored. A good or bad thing? . . . let's find out.

I'm using neural/genetic machine learning to process season-to-date league results to find the best fit of correct home, draw, away predictions to then apply to the next set of fixtures. In order allow the algorithms enough relevant information and data it is thought necessary to employ only where a league season is half completed.

Other caveats include no leagues of more than 20 teams so as not to overwhelm the net (and take far longer to process). No leagues with too few teams either where teams may play each other more than twice ~ nothing confuses artificial intelligence more than encountering the same situation with different outcomes, e.g. United v City showing one home win and one away win.

Having churned through a few leagues here's what it gives for this weekend; (odds are Oddsportal average odds from 50 odd bookies)

CHILE
18-Aug-18	16:00	San Luis	U. Espanola	A	2.57	3.28	2.59
18-Aug-18	19:00	Antofagasta	Union La Calera	nobet	1.94	3.54	3.57
18-Aug-18	21:30	U. De Chile	O'Higgins	A	1.64	4.01	4.51
19-Aug-18	00:00	Everton	Huachipato	H	2.04	3.27	3.52
19-Aug-18	16:00	Deportes Iquique	Colo Colo	A	3.19	3.51	2.05
19-Aug-18	19:00	U. De Concepcion	Deportes Temuco	H	1.83	3.43	3.99
19-Aug-18	21:30	U. Catolica	Curico Unido	H	1.72	3.57	4.45
SWEDEN
18-Aug-18	15:00	Malmo FF	Trelleborgs	H	1.15	7.89	17.16
19-Aug-18	14:00	Elfsborg	Norrkoping	X	2.87	3.44	2.36
19-Aug-18	14:00	Sundsvall	Hacken	H	3.26	3.51	2.13
19-Aug-18	16:30	Brommapojkarna	AIK	nobet	7.98	4.47	1.4
19-Aug-18	16:30	Djurgarden	Kalmar	H	1.65	3.63	5.53
20-Aug-18	18:00	Hammarby	Orebro	H	1.43	4.64	6.64
20-Aug-18	18:00	Goteborg	Ostersunds	A	2.93	3.25	2.41
JAPAN
19-Aug-18	05:00	Sapporo	FC Tokyo	X	2.79	3.27	2.45
19-Aug-18	10:00	Nagoya	Sagan Tosu	X	2.9	3.4	2.3
19-Aug-18	10:00	Shimizu	Urawa	H	3.54	3.38	2.03
19-Aug-18	10:00	V-Varen Nagasaki	C-Osaka	X	3.43	3.4	2.05
19-Aug-18	10:30	Kashima	Yokohama M.	A	1.92	3.77	3.53
19-Aug-18	11:00	Hiroshima	Kawasaki Frontale	A	2.88	3.41	2.32
19-Aug-18	11:00	Iwata	Kashiwa	A	2.74	3.17	2.55
19-Aug-18	11:00	Shonan	Kobe	X	3.23	3.42	2.13
19-Aug-18	11:00	Vegalta Sendai	G-Osaka	A	2.09	3.53	3.22
CHINA
18-Aug-18	12:35	Beijing Renhe	Beijing Guoan	A	7.35	5.74	1.31
18-Aug-18	12:35	Chongqing Lifan	Shanghai SIPG	A	7.94	5.83	1.29
18-Aug-18	12:35	Dalian Yifang F.C.	Henan Jianye	H	1.91	3.48	3.77
18-Aug-18	12:35	Guangzhou R&F	Changchun Yatai	A	2.14	3.62	3.01
18-Aug-18	12:35	Guizhou Zhicheng	Shandong Luneng	A	6.24	4.93	1.4
18-Aug-18	12:35	Tianjin Teda	Jiangsu Suning	nobet	2.43	3.38	2.71
19-Aug-18	12:35	Guangzhou Evergrande	Tianjin Quanjian	H	1.07	10.81	22.57
19-Aug-18	12:35	Hebei	Shanghai Shenhua	H	2.09	3.72	3.01

I also processed Brazil Serie A, but they're not quite half way through their season yet, but here are the forecasts anyway;

BRAZIL					OddsPortal	average	odds
18-Aug-18	20:00	Santos	Sport Recife	H	1.51	4.01	6.46
18-Aug-18	23:00	Corinthians	Gremio	A	2.97	2.92	2.58
19-Aug-18	15:00	Atletico-PR	Flamengo RJ	H	2.74	3.12	2.62
19-Aug-18	15:00	Internacional	Parana	H	1.38	4.4	8.74
19-Aug-18	20:00	Botafogo RJ	Atletico-MG	X	2.36	3.31	2.92
19-Aug-18	20:00	Cruzeiro	Bahia	H	1.44	3.98	8.2
19-Aug-18	20:00	Vitoria	Palmeiras	X	3.42	3.4	2.08
19-Aug-18	23:00	America MG	Fluminense	H	2.31	3.09	3.19
19-Aug-18	23:00	Sao Paulo	Chapecoense-SC	H	1.39	4.31	8.88
21-Aug-18	00:00	Vasco	Ceara	A	1.75	3.36	4.91

 

Edited August 18, 2018 by Data
Info

[Data]

14 correct from the 38 picks (including the Brazil games) is not a start I would have wished for. -12.83 from a 33.8% strike rate and a -33.76% yield. But this thread is experimental and my thinking-in-progress, so all is far from doom & gloom.

My neural net classifier gives a binary decision on each game. For my first run, with last weekend's fixtures I merely selected the outcome with the highest network 'probability' for every match, with a no bet where two oucomes were too close to call. Better I think to select only the very best of the picks where the net is suggesting a near certainty. I'm not naive enough to think such games will be anywhere near the certainty suggested, but they will be the cream of the crop.

Elsewhere I've expressed my opinion that football draws are probably the hardest outcome to predict, and my weekend sample, although comparatively small, suggested 7 such outcomes - and every single one failed! Not a proof by any means, but something to keep in my mind.

So, the Brazil net newly trained with the latest results and bearing the above thoughts in mind, we have three midweek selections; ALL HOME WINS:-

BRAZIL A
23-Aug-18	01:00	Palmeiras	Botafogo RJ	1.44
23-Aug-18	01:00	Sport Recife	America MG	1.91
23-Aug-18	01:45	Fluminense	Corinthians	2.23