Master’s thesis carried out at Emerse Sverige AB for the Department of Computer Science, Lund University.
Authors: Carl Dahl, Pontus Ericsson.
Pierre Nugues, Department of Computer Science, Faculty of Engineering, Lund University
Jacek Malec, Department of Computer Science, Faculty of Engineering, Lund University
Carl-Johan Grund, Emerse Sverige AB
Rasmus Larsson, Emerse Sverige AB
Articles in this series are theoretical and involves a substantial part of mathematics and computer science.
This thesis is an exploratory study into the possibility of using machine learning to manage advertisement campaigns and agents involved in real-time bidding. The norm for the industry of real time bidding is currently having human operators managing campaigns by changing settings to maximize the number of clicks. The goal was to investigate the possibility of automating this process, to at the very least assist the human operators with making better decisions. The first part of the project was to build a model for predicting the clickthrough rate (CTR) of the ad campaigns. The second part was to use the model to suggests optimal settings for bidding agents. The outcome was a model with an accuracy of 92% in predicting whether an ad was to generate any clicks or not, and with an accuracy of 58% to predict the outcome of an agent in the different categories “few clicks”, “some clicks” and “many clicks”.