We explored methods of improving upon Chitika, Inc.'s existing means of predicting which users would most probably click on an advertisement in a mobile application. We used machine learning algorithms, primarily Naive Bayes, that trained on demographic and behavioral information supplied by the user and his/her mobile device. After an exploratory phase, we gathered performance data using the AUC metric on twenty-eight different experimental conditions. When compared to the control condition, in which no preprocessing was performed on the data before being given to the unmodified Naive Bayes algorithm, we found only minor improvements in AUC.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.