Improving Customer Churn Prediction by Data Augmentation Using Pictorial Stimulus-Choice Data

The purpose of this paper is to determine the added value of pictorial stimulus-choice data in customer churn prediction. Using Random Forests and 5 times 2 fold cross-validation, this study analyzes how much pictorial stimulus choice data and survey data increase the AUC of a churn model over and above administrative, operational and complaints data. The finding is that pictorial-stimulus choice data significantly increases AUC of models with administrative and operational data. The practical implication of this finding is that companies should start considering mining pictorial data from social media sites (e.g. Pinterest), in order to augment their internal customer database. This study is original in that it is the first that assesses the added value of pictorial stimulus-choice data in predictive models. This is important because more and more social media websites are focusing on pictures.