ADVICE NEEDED, JAVA SPORTS PREDICTION / FORECASTING ALGORITHM

Question

Hi,
I am considering programming a Java Application which predicts sporting results (based on previous results),
Which forecasting algorithms should I use to predict/forecast results of future matches based on previous results? (or how should I go about this task)
Any feedback or ideas are very appreciated.

Answer 1

Any forcasting method whatsoever, is based on examining previous history, assigning relavance to previous event (like the team lost because it was away from home, or they don't play well in the rain) in order to make the prediction.
So you must create a model of what historical events you believe to be important, (this is essentially a free choice on your part) and then you collect that history, assign the relevance to each event, (using whatever relevance you want to attach to each chunk) and then you use what ever algorithm you like to combine the results and make your prediction.
Of course, you model has very litte to do with reality, and depending on how much work you do in designing your model, reality may or may not be related to your model.
So your question about what forcasting method to use is somewhat irrelevant. The real qestion is how much information do you have access to, how much work do you want to do etc.
Consider for example the following: Suppose you had asked the exact same question not about sports teams game outcomes, but rather about the outcomes of a coin flipping. You want to write a Java application that will tell me which way the coin will land, Heads or Tails. What is the right algorithm to use? baysian analysis, time series analysis, etc.
Hopefully, you can see that it just doesn't matter. Since the event is the classical example of an unbiased random event, your predictions will be right half of the time and wrong half of the time. You can use a random number generator to predict heads half of the time and tails half of the time, or you could write a much simpler program and just predict Heads every single time. Simple code and just as good as anything more complicated.
You like to believe that the outcome of a sporting event is NOT a classical random event and that you ought to be able to do a better job at predicting than just guessing. The only way you will know if you can do a better job consists of creating theories about what data might be relevant to improving your guesses, building models based on those theories and seeing if the model does better than random guessing. This is what statisticians do.
The poster that suggested that you use a random number generator is not far from wrong. That was simply proposing the theory that the process is random, not predictable and therfore not worthy of spending time writing extra code. That model completely ignores historical evidence and generates a result and that model is in fact a predictor and will be correct a certain percentage of the time. (of course, a simpler predictor that does nearly the same is to just predict a win every single time, shorter faster code)
In the weather business, where they look at weather patterns and satelite photos, and study the entrails of goats, there is a common simple model. It is good. It is better than random guessing. It is the gold standard of weather predicttion. All other prediction systems are measured against this model. The model is simply this. Predict for tomorrow the weather you see today. If it is raining today predict that it will rain tomorrow. If it is sunny predict sunny.
You are correct on any continuation of previous weather and you miss any change and guess what, continuation is more common than change. This model works so well, and furthermore is so trivial, that anyone that is trying to sell a more complicated model of the weather must first show that the new improved model predicts better than the trivial model, because if it is only that good, forget it.
So my advice to you is do exactly the same for you teams. Just predict that they will do the same as they did last time.
First of all, this is a simple model and easy to implement.
Second it forces you to think about how you get access to historical information and how you can incorporate it into a decision process.
Third it will probably do the same as the weather. Good teams that win will tend to win, loser will tend to lose.
You can get all the other components of this system working allowing this to be a place holder that you can always replace later when you discover the superior predictor, if there is one.
However I will make a prediction. You will have to work pretty hard to do better than this simple algorithm and given that you asked this question you have already indicated that you don't know much about prediction and forecasting algorithms and thus it is unlikely that you will make much headway on this difficult problem.
There is on the other hand a good chance that if you do everything else right, build an nice user interface, and make your code easy and fun to use, someone will come along with some suggestions about how you could improve your predictions and then the statistical model building fun can begin. Just make them first prove that their more complicated predictor works better than the gold standard, "What happened today will happen tomorrow"
Enjoy