Predicting the Level of Sustainability in Bike-Sharing Systems Using Machine Learning Techniques: A Study on RideMovies in Vicenza

This study explores the use of machine learning to predict the sustainability levels of users in bike-sharing systems, using data from the RideMovi service in Vicenza, Italy, collected between 2022 and 2023. To capture the multifaceted nature of sustainability, we defined key performance indicators (KPIs) related to environmental, social, and economic dimensions, based on data such as trip duration, distance traveled, promotions utilized, and payment. These KPIs provided a structured way to calculate sustainability scores for each user, which were then classified into two levels: very low and low-medium. Three machine learning algorithms—Generalized Linear Models (GLM), Random Forests (RF), and Support Vector Machines (SVM)—were applied to predict users’ sustainability levels based on these calculated attributes. To better understand the impact of individual variables on model predictions, the Explainer function was conducted. This analysis revealed that different variables were significant for each model: payment was the most influential variable for the Rf and GLM models and duration for the SVM model. Model evaluation demonstrated that RF achieved the highest accuracy (98.66%), followed by SVM (97.33%) and GLM (96.23%). Moreover, RF exhibited the highest Kappa value (0.941), indicating superior consistency in classification beyond simple accuracy. These results highlight the potential of data-driven methods, particularly machine learning, in enhancing our understanding of user behavior within urban mobility contexts. The insights gained from this research can be valuable for policymakers and city planners, providing guidance on promoting sustainable practices and improving the overall sustainability of urban transportation systems.