Attribution modelling of online advertising

Attribution modelling is one of the most sought-after research topics in digital marketing. While much research and progress has been made into predictive modelling, attribution modelling requires abundant domain expertise and interpretability for it to be adopted and used by marketers. Many approaches have been laid out including logistic regression and graph-based attribution models such as markov chain which has shown consistent results while retaining high interpretability. In this thesis, we work on a data set which includes logs of user activity such as user clicks, impressions, and user conversions. The thesis makes use of two different kinds of analysis, user level and sequence level in which different logistic regression models and markov chains are used to assess the performance of attribution on a varied set of metrics and address the class imbalance problem which frequently occurs with user log data