The main round of the World T20 tournament is currently underway in Bangladesh. During this tournament, we will generate a prediction before each match as to which team is most likely to win that match. The prediction for the match(es) on each day will be published on the website in a blog post.
One match in the main round has already been played, in which India easily defeated Pakistan. However, we had predicted just the opposite outcome – that Pakistan has 66% chance of defeating India. Nevertheless, it is just one match, and it is impossible for any prediction system to be 100% correct all the time. We are hopeful that our methodology will yield more accurate predictions in the coming matches.
The main algorithm behind match prediction is quite similar to the one that we had used for predicting the outcome of the IPL matches last year. The main idea is to assign a score to each player of a team, based on their past performances. The score of a team is the total score of the probable 11 players of that team that will play the upcoming match. The chance of a team winning a match against an opponent is the ratio of its own score to the sum of the score of the two teams.
For predicting IPL matches, we had used the Runs Above Average (RAA) statistic for calculating the score of a player. For the World T20 tournament, we will instead use the EV score for the player score. The advantage of EV score is that it takes into account the quality of opponents a player has faced when calculating his score, and therefore better reflects the strength of that player. It also encapsulates both the batting as well as the bowling score nicely into one single number. The EV scores are calculated based on all the T20I matches played in year 2012 and on wards. As the tournament progresses, we will use the data of the most recently played matches as well when computing the player scores. So hopefully, our predictions will improve during the course of the tournament.
You can follow the pre-match predictions on this blog, or on our Twitter/Facebook/Google+ page.