Predicting Preferences of Voters from Opinion Polls by Machine Learning and Game Theory

We present a method to determine political profiles of a voters population through the application of learning algorithms and game theory. Our process began with a collection of surveys gathered in two ways, from a website and directly from voters. Having a linear hierarchy on the attributes expressing the political preferences of the voters allow to apply a zero-sum game approach to distribute the potential number of votes among the involved candidates. We have applied our model to do electoral prospection in a case study of the Major’s election of our city, that took place in July 2013. The results were quite encouraging. Our approach has also the potential to be applied to new product advertisement campaigns.