Predicting website exits with machine learning

Sammanfattning: Website hosts want interested visitors who engage in the activities that are the purpose of a website. This is usually achieved by designing the website to be simple to navigate and aesthetically pleasing. As the design is done before the visits, it would be an exciting addition if the website could identify the visitors who become disinterested during their visits, and then offer personalized motivation for the visitor to engage in the website. This study aims to identify whether a visitor is about to leave a website, by using machine learning models to predict their exits. The purpose of that is to offer personalized motivation in real time for visitors to continue their visits. That is however outside the scope of this study. This research investigates how well machine learning models can predict website exits from session data. The algorithms chosen for the prediction are an Artificial Neural Network (ANN) and a Support Vector Machine (SVM). These are trained on session data. An important part of the research is to extract suitable features from the session data to enhance the prediction. The models are cross-validated and it is found that the models show success in predicting exits, and that it is possible to predict exits with a performance of at least 0.70 Area Under the Curve (AUC).