Prediction of Payment Method in Carsharing

Abstract

The rise of urban mobility platforms using VTC licenses (Uber, Cabify, Bolt) allows users to pay in cash or via registered cards/accounts. This study applies machine learning to predict the payment method (card or cash) based on income levels. After cleaning incomplete records, we compiled 280,000 VTC services, focusing on “origin postal code” and “payment method,” alongside “disposable average income” from the Tax Agency's IRPF statistics. Results show 85% accuracy, with a training accuracy of 0.865 and test accuracy of 0.808. Precision reached 0.841 (train) and 0.792 (test), recall 0.885 (train) and 0.843 (test), and MCC 0.714 (train) and 0.617 (test). Findings confirm that lower-income postal codes correlate with higher cash payments due to financial exclusion, including lack of banking access for irregular immigrants, difficulties in obtaining credit cards for low-income individuals, and account garnishments promoting informal economy reliance.