Machine Learning Methods for Click Fraud Detection

Online advertising is the fastest-growing sector in the advertising industry, and in some countries, online advertising already accounts for more than half of total advertising expenditure. Various reports have shown that the damage caused by advertising fraud has amounted to several billion dollars in recent years. These sums show that there is a need for action to curb fraud and reduce the losses for advertisers. The goal is to evaluate whether supervised machine learning methods are able to support click fraud detection and thereby prevent fraud and reduce financial losses. To achieve this, the online advertising ecosystem and its weak points for click fraud will first be examined by means of literature analysis. Based on the literature, possible supervised machine learning algorithms for the practical part of the work will be identified. Afterwards, explorative data analysis will be performed and the data will be prepared before models for the selected algorithms are created and trained. Finally, the performance of the trained models is evaluated and the results of the different models are compared. The results of this work is, that supervised machine learning models are able to support detecting click fraud. The classification works well but not all observations are correctly marked. Therefore, supervised machine learning models can be used to mark or pre-filter questionable actions, which are then checked by a human operator.