Neural Networks (revisited)

[PAULM03]

I've been thinking about this for a while now, and I finally have a small amount of time to commit so I'd like to try to get something going again. Obviously many heads are better than one so any input would be appreciated. This is where I'm at at the minute: I'm getting to grips with this program, which to my limited knowledge on the subject looks the bees knees: http://www.jooneworld.com/ Very user friendly so far and I've been playing around doing the tutorials and running some imaginary data of my own just to get used to the interface. I feel like I'm now in a position to use the program on some historical football data, and obviously this is where it gets interesting (and slightly tricky) as there are 2 issues that I can see: 1) Selection of suitable input and output variables 2) Manipulation of data into a suitable form for processing Starting with the first one: I suppose it's good to start with a simple aim, so to start with I think I'm going to try to create a network to deal with the over/under market. This would give a simple binary output i.e. 1 for over 2.5 goals and 0 for under 2.5 goals So the next step is to decide which variables affect whether a match finishes over or under. The obvious variables to start with would be the relative goalscoring and defensive records of the 2 sides, this would give 4 input variables: Home team goals scored, Away team goals scored, Home team goals conceded and Away team goals conceded. This could also be split further into home goals scored, away goals scored etc. etc. to give 8 variables. The next step is where I'm having difficulty, I think the best way to use these variables would be to look at each sides attacking and defensive capability relative to the rest of the league, that way you can give each variable a value between 0 and 1. As an example: Say Chelsea are the top scorers in the division and are playing at home then the value of the 'Home team goals scored' input variable would be 1 And say they were playing Aston Villa who were the lowest scorers in the division then the 'Away team goals scored' input variable would be 0. However I'm finding it incredibly tricky to think of a way I can manipulate the data from football-data.co.uk to give me these values for each team in each game, and this is where I really need some input from someone who has experience of manipulating reasonably large data sets in excel or some other program. Does anyone have any ideas? Hopefully there's some other S&S posters with an interest who would be interested in contributing and developing some ideas and hopefully making a reasonable attempt at applying an NN. Cheers:ok

[cavello]

Re: Neural Networks (revisited) Suggested inputs for under/over. Price of stronger team for a win. The bigger the different in ability of teams the more likely of more goals. ELO rating differences. Average goals scored by stronger team, average goals conceded by weaker team. Average number of goals for the League. How you weight the inpust initially is up to you.