Modelling In-Store Consumer Behaviour Using Machine Learning and Digital Signage Audience Measurement Data

Abstract

Audience adaptive digital signage is a new emerging technology, where public broadcasting displays adapt their content to the audience demographic and temporal features. The collected audience measurement data can be used as a unique basis for statistical analysis of viewing patterns, interactive display applications and also for further research and observer modelling. Here, we use machine learning methods on real-world digital signage viewership data to predict consumer behaviour in a retail environment, especially oriented towards the purchase decision process and the roles in purchasing situations. A case study is performed on data from a small retail shop where demographic and audience data of 1294 store customers were collected, manually verified and analysed. Among all customers, 246 store customers were involved in a buying process that resulted in an actual purchase. Comparison of different machine learning methods shows that by using support vector machines we can predict with 88.6 % classification accuracy whether a customer will actually make a purchase, which outperforms classification accuracy of a baseline (majority) classifier by 7.5 %. A similar approach can also be used to predict the roles of an individual in the purchase decision process. We show that by extending the audience measurement dataset with additional heuristic features, the support vector machines classifier on average improves the classification accuracy of a baseline classifier by 15 %.