Predicting customers' cross-buying decisions: a two-stage machine learning approach

Predicting a customer's cross-buying behaviour is a challenging problem for many organisations. In this paper, we propose a novel two-stage cross-buying prediction framework by integrating machine learning, feature engineering, and interpretation techniques. Specifically, the first stage aims to train an accurate complex black-box classification model with cross-validation and hyperparameter tuning. Then, the next stage uses the top ten most important predictors of the black-box model to obtain a simple rule-based interpretable model. We use a publicly available dataset published on the Harvard Dataverse to provide a practical case study. The results show that the rule-based model has a predictive performance as high as the complex model.