The purpose of the diploma work was to gain new knowledge about the methods of advertising, the characteristics of online campaigns and to predict the number of campaign impressions after a certain period of duration. In the first part of the thesis, we built a data set from different sources and analyzed the data according to different campaign characteristics (number of impressions, year of activity, duration). We discovered different patterns, especially in terms of campaign duration. In the second part, we tested the performance of predicting the number of campaign impressions. We used three data sets that differed by the time we started to collect forecasting data. We used 5-fold cross-validation to evaluate five regression methods (linear regression, regression tree, random forests, support vector machine, k nearest neighbors) for the task.
|
