View Single Post
Old December 3, 2012, 07:28 AM
Zunaid Zunaid is offline
Join Date: January 22, 2004
Posts: 22,099

Originally Posted by zsayeed
Last 10 year BD has won 41% in Dhaka, last 5 years: 39%, last 2 years won 29%, and last 12 months won 40% (exactly) of the matches. Probabilistically speaking BD's probability of win is 40% I would think, but I feel it will be close to 80% by recent performance with this WI team. In fact - in the 1st test we almost should have won.
That's not quite a valid assertion because the assumption being made here is that all opponents are of equal strength. For a better predictive power, we will need to factor in the strength of the opposition at the very least.

Looks like a simple machine learning model could be constructed. For simplicity, we can start with the following features:
  • Oponent strength - we can use ICC rating as a value
  • Type of match (ODI/TEST/T20I)
  • Ground
Of course, we can start more and more features such as month of match, who batted first, team compositions (how many spinners vs pacers), etc etc.

Of course, the more features we have, we need to come up with some feature selection algorithm - or we may be in danger over fitting.

I just have a gut feeling that we might get best results with an SVM. But YMMV.
Reply With Quote