# How would I go about analysing MMA statistics and data for future predictions?

My maths knowledge is pretty poor but I'm looking to improve on a new project I'm interested in.

I have data from thousands of MMA fights/fighters including weights, height, reach etc. and I'm trying to make a calculation to predict future fights.

Now I know this is probably a huge task, but I'm thinking of it more as an exercise of how I'd go about doing it, rather than expecting accurate results.

How would I go about analyzing data like this? What should I be reading into, and what methods should I be using to get to find patterns in the data etc.? Any resources to start me off would be hugely appreciated.

I can remember very little of mathematics from school, so the more simple the better.

• Interesting that we cannot assume that you know anything much, but you assume we know what MMA means. Googling suggests "mixed martial arts". Commented Jun 19, 2013 at 9:13
• @Hi john, welcome. There are numerous machine learning techniques for predictive modelling. Common methods include: decision trees, random forests, support vector machines (SVM), Baysian networks, neural networks, boosting etc. Googling each of those terms together with "introduction" or "tutorial" will yield several interesting results. Commented Jun 19, 2013 at 9:36
• @COOLserdash is here I think being optimistic that you can ingest much of modern machine learning just like that by reading online. Students taking courses and with college-level mathematics background expect to take a long time to learn this stuff. Commented Jun 19, 2013 at 9:57
• @john Have you looked into prediction methods used in other sports/games? For example, Zermelo's (1929) model can be used to model the strength of two chess players (more notes on this from University of Cincinnati). Chess has a rating system and supposedly this is a good indicator of which players are better. MMA does not have such a system (?), but an option could be to try and come up with your own one and form predictions based on that. Commented Jun 19, 2013 at 21:19
• (cont...) In the past I've come across posts on thesportseconomist.com that touch upon the sort of thing you're trying to do. The search function on the website isn't great, but you might hit on something there. Try searching the Journal of Sports Economics too. Commented Jun 19, 2013 at 21:23

Predicting something of interest from available data is arguably the central problem of modern statistics, so there is plenty.

But you really shouldn't expect to hit the ground running. You should expect to work through a good introduction to statistics with a strong emphasis on correlation and regression, which in practice means almost any of them. I like Freedman, D., Pisani, R., Purves, R. Statistics (any edition), although plenty of people don't.

The big blockbuster texts aimed at introductory students (Agresti and Frankin; De Veaux, Velleman and Bock, etc., etc.) get as far as multiple regression, which is important for you.

It's fairly large undertaking however it's not so hard to get started. I've written a short paper which addresses methodological concerns for beginners here:

http://prescientmuse.blogspot.co.uk/2015/01/mixed-martial-arts-fight-outcome.html

It'll give you a good idea of what sort of things are involved. Further, here is a list of stuff not to assume:

http://prescientmuse.blogspot.co.uk/2015/01/assumptions-to-avoid-when-predicting-mma.html

Otherwise, consider googling for introduction to experimental design and analysis. Let the statistical knowledge increase in proportion to your knowledge of how to actually design and implement experiments. There are many wrong ways to apply statistics for each right one.

Let me know if you have more specific questions.