Predicting truck driver churn in online mutual help communities with penalized survival models

Purpose This study investigates user churn in online mutual help communities (OMHCs) for truck drivers, identifying when and why disengagement occurs to enable early warning and targeted retention strategies. Design/methodology/approach Using large-scale behavioral logs from a truck-driver OMHC, this study models time-to-churn with Cox proportional hazards, estimates Kaplan-Meier survival curves across user groups and applies penalized variable selection (Lasso, Smoothly Clipped Absolute Deviation, Minimax Concave Penalty) to high-dimensional features capturing social capital, individual characteristics, and environmental factors. Model sparsity and predictive performance are compared across penalties. Findings Survival analysis reveals that retention declines over time, with pronounced heterogeneity across groups. Churn is slower for branch members, mutual-help insurance purchasers, “mutual-help pioneer” recipients, certified or license-verified users, higher point levels, and membership in higher-level branches. Penalized Cox models identify key factors influencing churn, with Lasso achieving the greatest predictive accuracy, MCP offering the most parsimonious model, and SCAD balancing accuracy and sparsity. Originality/value This study extends churn research to truck drivers' OMHCs by integrating community logs with external environmental data and structuring predictors into social capital, individual, and contextual domains. Methodologically, it advances survival analysis for high-dimensional community data through comparative penalized Cox frameworks, providing a replicable and interpretable approach for churn detection and retention optimization in road transportation-focused digital communities.