Artificial intelligence (AI) algorithms are increasingly used in ride-sharing to organize dynamic pairings, predict the most suitable drivers for passengers, rapidly optimize routes, minimize empty miles, and ensure safer and smoother rides. Thus, AI can assist in realizing a greener, smarter, and more equitable ride-sharing landscape. This article discusses several recent studies that demonstrated the benefits and effectiveness of using AI in ride-sharing and the challenges of using AI in this field.
Image credit: antoniodiaz/Shutterstock
Importance and Effectiveness of AI
Reduction of Social Barriers: Ride-sharing can substantially reduce traffic congestion, parking infrastructure, and per-passenger carbon emissions and ensure more cost-effective traveling compared to conventional solo ride-hailing. However, social barriers pose significant challenges in ride-sharing as riders do not possess any knowledge of other commuting riders.
In a paper published in the journal Procedia Computer Science, researchers proposed an enhanced ride-sharing model (ERSM) in which riders are matched based on a specific human characteristic set using machine learning (ML) to address the limitations of social barriers in the existing ride-sharing models.
The proposed model for ride-sharing was based on rider characteristics and an ML content-based recommendation system, with the broadcasting rider request, the closest driver, matching layers, the feedback system, and the ML module being key elements of the model.
In the model, the first task was identifying the five human characteristics on a 1 to 5 scale. The selected characteristics, including friendliness, comfortability, safety, chatty, and punctuality, are associated while commuting in a group/traveling on a trip. The task also included identifying an estimated/approximate maximum waiting time for a rider while picking other riders.
Riders were added to the trip itinerary if they satisfied the matching layer conditions. Additionally, the system also consisted of a newly designed feedback system that allowed riders to rate the driver and other riders based on the five identified characteristics on the designated scale.
Later, the proposed model used the feedback dataset to compute two classifiers for each user. These classifiers indicate the characteristics a user/rider expects in other users/riders while commuting on a trip. Eventually, every rider was associated with two major characteristics, and the system paired the riders with similar classifiers traveling on a similar trajectory. The computed classifiers and registered characteristics were then fed to an ML module to predict the newly registered riders’ characteristics. These predicted characteristics were used to recommend the riders in future trips.
Support vector machine (SVM) was selected as the ML prediction module. Two datasets, including Feedback-Received-Classifier ML-data and Feedback-Given-Classifier ML-data, were created to train the SVM. In both datasets, the input fields were registered user characteristics and the output was the computed classifier. Two separate SVM modules were used for two datasets that gave Feedback-Given-Classifier and Feedback-Received-Classifier outputs. The module was also evaluated with newly registered users.
Researchers performed extensive simulations to assess the system's performance. Both modules predicted the two classifiers for the newly registered riders, and the system recommended riders based on the predicted classifiers. The number of completed trips and the matching rate continued to rise with the progressing simulations/increasing number of traversed riders. Additionally, the SVM modules achieved 90% accuracy and accurately predicted classifiers for newly registering riders, which was crucial in providing real-time and better rider recommendations.
Moreover, the overall trip formation time was approximately one minute. These findings demonstrated the feasibility of using the proposed model to complete a maximum number of trips with pool completion and achieve maximum rider matches through closer and exact characteristics matching.
Identifying Factors that Influence Shared Rides: Although ride-sharing has several benefits, it accounts for only a small percentage of the total ride-hailing trips in cities. In a study published in the journal Travel Behaviour and Society, researchers developed an analytical framework that integrated big trip data with scalable ML and eXplainable AI (XAI) to identify the factors influencing willingness to take shared rides/factors associated with people’s decisions to take shared rides.
Although they used the City of Chicago as a case study in this work, the workflow can be replicated in other cities where trip data are available. The study primarily focused on identifying the complex non-linearities and interaction effects overlooked in earlier studies.
Extreme Gradient Boosting (XGBoost) was utilized as the ML model as it is one of the extensively used ML methods for supervised classification and regression tasks and outperforms other approaches, such as deep neural networks (DNNs) and random forest (RF), on tabular data. Additionally, SHapley Additive exPlanations (SHAP), a local interpretable ML method, was used to explain and attribute the XGBoost predictions to each feature.
The study results demonstrated that users adopt ridesharing for longer-distance trips and the trip cost remains the most critical factor. Researchers also identified a strong diurnal pattern in which users preferred to request shared trips during peak morning and afternoon hours.
Additionally, they observed that socio-economic disparities are crucial in influencing the willingness to share rides. For instance, private rides were preferred in more affluent neighborhoods as residents are less sensitive to price/ride-matching, which is less effective.
Thus, the XAI-based analytical framework proposed in this study can assist local governments and transportation network companies understand ridesharing behavior and recommend new policies and strategies to promote ride-sharing for efficient and sustainable transportation.
Optimal Dispatch of Available Vehicle Fleet: The success of modern ride-sharing platforms primarily depends on the ride-sharing fleet operating companies’ profit and efficient management of available resources. Customers waiting time and extra travel time, which are major performance metrics in ride-sharing, can be reduced by optimally dispatching the vehicles to various locations in a service area.
In a study published in IEEE Transactions on Intelligent Transportation Systems, researchers developed a distributed model-free algorithm DeepPool that utilizes deep Q-network (DQN) techniques to learn optimal dispatch policies by interacting with the environment. DeepPool can incorporate deep learning models and travel demand statistics to efficiently manage dispatching vehicles for better ride-sharing services in large-scale systems.
An optimization problem was formulated where a dispatcher wants to minimize four objectives, including the demand and supply mismatch, the waiting time of customers and the total time vehicles required to traverse to serve future demands, the additional time spent by the customers due to ride-sharing, and the number of vehicles used to reduce traffic congestion and fuel consumption.
DeepPool demonstrated better performance compared to other strategies in the literature that do not dispatch vehicles to regions where demand is anticipated in the future on the real-world dataset of taxi trip records in New York City. Additionally, DeepPool rapidly adapted to dynamic environments as it was implemented in a distributed manner in which every vehicle could solve its DQN independently without coordination.
Limitations
The development of an AI model or an AI-based approach for ride-sharing is extremely complicated due to the need to create mechanical intelligence and properly understand human-based information.
Additionally, accurately interpreting the outcomes of AI models is difficult due to the black box limitation, which makes these techniques less reliable. Data bias and discrimination, safety concerns, privacy issues, and ethical considerations are the other major limitations of AI in ride-sharing.
Specifically, using biased data to train AI algorithms can perpetuate social inequalities in ride-sharing. For instance, biased algorithms can result in higher prices or longer wait times for specific demographics. Similarly, AI-powered ride-sharing platforms collect and analyze substantial amounts of rider data, which can increase concerns about user privacy violations and potential misuse of information. Thus, ensuring data privacy and security is essential before implementing AI-powered solutions.
In conclusion, AI can revolutionize ride-sharing by optimally dispatching vehicle fleets, reducing social barriers to ride-sharing, and recommending strategies to promote ride-sharing. The existing limitations of using AI must be addressed to exploit the full potential of the technology.
References and Further Reading
Li, Z. (2023). Leveraging explainable artificial intelligence and big trip data to understand factors influencing willingness to ridesharing. Travel Behaviour and Society, 31, 284-294. https://doi.org/10.1016/j.tbs.2022.12.006
Yatnalkar, G., Narman, H. S., & Malik, H. (2019). An Enhanced Ride Sharing Model Based on Human Characteristics and Machine Learning Recommender System. Procedia Computer Science, 170, 626-633. https://doi.org/10.1016/j.procs.2020.03.135
Abduljabbar, R., Dia, H., Liyanage, S., & Bagloee, S. A. (2018). Applications of Artificial Intelligence in Transport: An Overview. Sustainability, 11(1), 189. https://doi.org/10.3390/su11010189
Al-Abbasi, A. O., Ghosh, A., Aggarwal, V. (2019). DeepPool: Distributed Model-Free Algorithm for Ride-Sharing Using Deep Reinforcement Learning. IEEE Transactions on Intelligent Transportation Systems, 20, 12, 4714-4727. https://doi.org/10.1109/TITS.2019.2931830.