We present first results from a study of the customer interaction with an Internet travel booking platform. The focus of this study is to get a suitable and reliable prediction of the customer churn. Based on a pre-filtering of the big amount of raw data, we evaluate the individual customer sessions in order to come up with a segmentation of the customers. Criteria for this customer segmentation are for instance the session acitivity, ID coming from marketing campaings and the search behavior throughout the session and the booking funnel. As a basis for predicting the customer churn we discuss several possible definitions of churn on Internet travel booking platforms. As predition approaches we focus on a Poisson-based SMUCE estimation and a Bayesian approach and present first results for the underlying data. The results will be used in the research project OCIDA (\“Optimizing online customer interaction by advanced data analytics\“) with the goal of optimizing marketing strategies for churn reduction in terms of return of investment. We close with giving an outlook on a combined prediction and optimization approach.
