Rider churn prediction model for ride-hailing service: A machine learning approach

Abstract

This work attempts to propose a machine learning approach for rider churn prediction model in 4-wheel ride hailing industry based on the transactional data of Uber Peru 2010. Over the last decade there has been increasing interest for relevant studies in non-contractual churn prediction (e.g., telecom, assurance, banking). In contrast, there are extremely less researchers to investigate the contractual churn prediction such as ride-hailing service. Thus, the research findings of this work provide an important guide for on-demand mobility companies to improve customer adhesiveness. Three machine learning algorithms have been applied including SVM, Logistic Regression and Gaussian Naïve Bayes. The experiments were performed in Google Colab with the aid of Python programming. The results showed that all classifiers could achieve a greater-than-90% accuracy, thus applying any one can faily obtain good results in pratice. Depending upon the characteristics of ride-hailing industry and the measurements (Accuracy, Precision, Recall), Logistic Regression was chosen as a best binary classification model to predict whether or not a rider will churn within 3-month threshold.In order to effectively identify different types of churners for better retention solutions, values of the churners were divided into three segments using RFM theory. The results revealed that the likelihood of churn tendency is significantly reduced as the frequency and the monetary increase.