Optimizing pre-pickup ride retention in ride-hailing platforms using behavioral features with reinforcement learning

Ride-hailing services frequently encounter cancellations during the crucial time window between ride requests and rider pickups, impacting the platform’s efficiency and service quality. This study introduces a Reinforcement Learning (RL) based methodology to minimize such pre-pickup ride cancellations by dynamically adjusting incentives and fares at the time of the ride request. The framework excludes post-trip data like trip duration or revenue, focusing strictly on the pre-pickup phase. To simulate natural decision-making, synthetic behavioral features like Rider Patience Index (RPI), Driver Patience Index (DPI), Cancellation Rate (CR) and Historical Demand Forecast (HDF) are engineered by modeling time and contextual inputs. Utilizing an 8-dimensional state space, a Proximal Policy Optimization (PPO) agent is trained to output incentive values and fare adjustments. The proposed system is assessed with reference to average cost, simulated platform efficacy, and completion rate. Findings showed the RL agent outperformed the baseline rule-based approach with a more cost-efficient and adaptive solution for minimizing cancellations on ride-hailing platforms. The trained agent achieved 1.22% less platform cost compared to the baseline approach for completed rides showcasing superior performance.