Predicting Bike-Sharing Adoption for Green Mobility Planning Using Hard Voting Ensemble Classifier in Machine Learning: Evidence from Kigali City, Rwanda

Abstract Bike-sharing systems have proven to support sustainable urban mobility across the globe; however, limited research has integrated demographic, socioeconomic, and environmental factors to predict user adoption effectively. This study addresses this gap by applying a hard voting classifier that combines XGBoost, AdaBoost, and Random Forest models to analyze data collected from 5,000 residents in Kigali, Rwanda, enriched with corresponding air quality and weather variables. The model was evaluated using 10-fold cross-validation, and the voting classifier outperformed individual models with an accuracy of 97% (±0.5%), compared to 96% (±4%) for Random Forest, 96% for AdaBoost, and 93% for XGBoost. Feature importance analysis revealed that monthly income (13.3%), age (13.1%), location (11.9%), and household size (8.4%) were the most significant predictors of bike-sharing adoption. Among marital status groups, divorced individuals (4%) were more likely to adopt bike-sharing due to its flexibility in meeting independent travel needs. Environmental variables such as temperature (2.6%), humidity (1.5%), nitrogen dioxide (3.5%), and carbon monoxide (3.1%) also influenced usage patterns. These findings provide actionable insights for urban planners and policymakers to develop data-driven strategies that promote active mobility and reduce environmental pollution.