Abstract
This study bridges a gap in existing research by fusing behavioural variables from historical transactional records with demographic and psychographic data, aiming to predict systematic purchase behaviour (SPB). The research also investigates to what extent each of the group of variables helps understand the anatomy of SPB.
Data from over 2,481 stores, covering January 2012 to November 2015, was obtained from a leading UK grocery and pharmacy chain. Pseudo-anonymized records merged from transactional and survey data resulted in a cohort of 12,137 participants meeting the inclusion criteria. A novel measure, Bundle Entropy, derived from transactional data, evaluates SPB. Machine learning models and techniques are employed for predictive and variable importance analysis.
Preliminary results show that all variables in combination have a robust performance over random choices at predicting SPB, with RF and XGB being equally well-performing. SHAP and MCR results also show that behavioural variables have a more dominant role than socio-demographic, personality and psychological variables when predicting SPB. For instance, the significance of 'Total visits' suggests that frequent customer engagement serves as a key determinant, while the nuanced roles of age and variety-seeking behaviour underscore the multifaceted nature of consumer SPB.
Future research will delve into ML models relying solely on demographic and psychographic variables to discern their distinct contributions to SPB prediction.