Vision Transformer Approach to Customer Churn Prediction Radar Chart Image Classification for Non-subscription Based E-commerce

Having a ready-to-use dependent churn variable in the non-subscription-based e-commerce domain is not straightforward, as customers do not formally cancel their status as a customer but simply stop placing orders. In this study, a k-means clustering algorithm with RFM variable segmentation is used to assist in classifying customers as churned or not churned. Next, a novel approach is adopted by transforming tabular customer data into radar chart images which are then fed into a pre-trained Vision Transformer model for supervised image classification. The results of the study indicate that while the Vision Transformer model show significant increase in precision and F1-score evaluation metrics, its performance in MCC, recall, and AUCROC is comparable to that of CatBoost and XGBoost. This study demonstrates the potential of transforming tabular data into images for training Vision Transformers, for use in customer churn prediction within the non-subscription-based e-commerce domain.