Machine(s) learning to predict Super Bowl Winner!
Super Bowl is not for another 21 days and 21 hours but the machine learning algorithms already started to come down with predictions. The availability of extensive online data makes it possible to build such algorithms.
One such model was released by WSO2, an open-source service oriented architecture company with deep roots in Apache community. They used one of their services called Machine Learner (which uses Apache Spark under the hood) to build a random forest regression.
They picked their training strategy by using data from ’12, ’13, and ’14 to predict ’15, and compare it with the real results. This is similar to time series prediction. They got their historical data from http://www.pro-football-reference.com/.
Their model predicts that it will be North Carolina Panthers against New England Patriots with 76.5% accuracy. And their prediction for today’s game seemed to realize perfectly, Patriots beat Chief. You should check out their site at http://wso2.com/landing/big-data-game/:
It would be an interesting challenge to build a model with better accuracy. Unfortunately, WSO2 does not provide the data they used for training on their site, so any such effort will have to bootstrap the training data.
References:
- PCWorld article, 1/2016
