An evaluation of statistical models for programmatic TV bid clearance predictions

Abstract

This thesis presents three different classes of statistical models which are applied to predict bid clearance probabilities in first price auctions. The data set used comprise in two years’ worth of data from programmatic TV auctions where each bid has assigned class labels. Generalized Linear Models, Neural Networks and Support Vector Machines are described in detail together with methods for data pre-processing. Special focus is on Neural Networks and in particular on the comparison of the performance of different optimization methods which are evaluated in order to find the method most suitable. Findings indicate that Neural Networks perform on par, or better than the other methods mainly due to their high accuracy when predicting losing bids.