Title: Dynamic prediction of propensity to purchase by landmark modelling
Authors: Ilan Fridman Rojas - Profusion Media Ltd (United Kingdom) [presenting]
Aris Perperoglou - University of Essex (United Kingdom)
Berthold Lausen - Friedrich-Alexander-University of Erlangen-Nuremberg (Germany)
Henrik Nordmark - Profusion (United Kingdom)
Abstract: The aim is to present a novel application of a previous landmarking methodology to predict propensity to purchase based on transactional data. This use case presents a number of challenges, including data sets of considerable size for which many current statistical models and tools no longer scale to, and recurrent events with high frequency and multiplicity, often with time-varying covariates and strongly time-dependent effects. We present the results of such an application to subsets of a data set from a retailer with 2 million customers and 7 years of collected transactional data, extracting estimates of the time-varying effects, and producing dynamic predictions of probability of re-purchase which condition on time elapsed since the last purchase.