Supervised Aggregation of Classifiers using Artificial Prediction Markets Input ( x , y )

Abstract

Prediction markets are used in real life to predict outcomes of interest such as presidential elections. In this work we introduce a mathematical theory for Arti cial Prediction Markets for supervised classi er aggregation and probability estimation. We introduce the arti cial prediction market as a novel way to aggregate classi ers. We derive the market equations to enforce total budget conservation, show the market price uniqueness and give e cient algorithms for computing it. We show how to train the market participants by updating their budgets using training examples. We introduce classi er specialization as a new di erentiating characteristic between classi ers. Finally, we present experiments using random decision rules as specialized classi ers and show that the prediction market consistently outperforms Random Forest on real and synthetic data of varying degrees of di culty.

Cite this document ^(BETA)