Crickytics – Applying machine learning to Cricket

crickytics.com

35 points·Faizann20··9 comments

I have been playing with some cricket datasets in my spare time. You should document your process. Also, make some comparisons with the simplest regression model. Are you publishing the scraped datasets?

I did try linear regression but the results were bad. That's why I moved towards sophisticated models.

Interesting stuff, but does this actually perform better than using simple multiple linear regression on the same data?

Yes it does.

It's not necessary to apply sophisticated models like random forest/neural nets to cricket. It's a sports for which humongous amount of data is kept meticulously; also we're trying to predict pretty standard stuff like who'll win the next match. Standard regression should be able to perform very well (and it does).

I agree. Just cricinfo statusguru is a huge treasure trove, in addition to several ball by ball datasets from games.

Any links to something interesting to read that you suggest?

ieeexplore.ieee.org/document/7489605

Who will win the match? Sure.

But I think the prognostic value (monetary?) will be in ball-by-ball prediction/over prediction/no. of runs scored by a particular batsman, in which we will need a model that can be updated and the ability to ingest and make sense of voluminous amount of data. Some kind of data assimilation component too, maybe?

So I did try simple models but the results with sophisticated models have come out to be better than simpler ones.

Reply on news.ycombinator.com