Optimizing dynamic ridesharing systems: Enhancements in crowdsourced data quality and vehicle assessment algorithms

Abstract

This PhD thesis seeks to advance dynamic ridesharing systems by enhancing the quality of crowdsourced data in partially objective tasks and optimizing vehicle assessment algorithms. The research addresses key challenges within Intelligent Transportation Systems (ITS), focusing on two critical areas: refining lane change evaluations through improved crowdsourced data processing and optimizing ride-sharing recommendations to enhance carpooling services. The research adopts a dual approach. The first approach involves the development of an algorithm specifically designed to process and filter crowdsourced lane change evaluations. This algorithm effectively addresses worker biases and detects malicious inputs by employing a combination of linear modeling and fuzzy logic techniques. The second approach presents a comprehensive ride-sharing recommendation framework that integrates real-time traffic data, user preferences, environmental impact considerations, and vehicle capacity utilization. This framework was rigorously tested through simulations using a dataset encompassing 10,000 urban rides. Results from the lane change evaluation algorithm showed a significant reduction in noise and an improvement in data quality, leading to a substantial decrease in root mean square error (RMSE) when compared to baseline methods. Similarly, the ride-sharing recommendation framework demonstrated superior performance in several key areas compared to existing carpooling recommendation models. It effectively reduced the total mileage required to transport all passengers, leading to lower overall fuel consumption and a corresponding decrease in passenger fares. Additionally, drivers experienced an increase in profits due to the optimized matching of passengers based on shared preferences and the system’s ability to balance capacity utilization. However, while the system excelled in these areas, there remains some room for improvement in the percentage of satisfied ride requests. Despite this, the ability of QE-Ride to incorporate eco-friendliness and safety into its recommendations sets it apart from other systems, making it a promising tool for enhancing the overall quality of the carpooling experience for both passengers and drivers. These findings highlight the critical role of data quality improvement and optimized vehicle assignment algorithms in dynamic ridesharing scenarios. The proposed solutions not only enhance the technical capabilities of dynamic ridesharing systems but also provide practical strategies for improving sustainability and operational efficiency. Future research could further refine these data processing techniques and explore their application across other areas of ITS.