Intelligent Distributed Customer Anticipation Approach for Taxi Routing Optimization

The advent of ubiquitous mobile computing technology has opened doors for innovative applications that effectively utilize information from IoT devices in a real-time and seamless manner, disrupting many traditional business procedures. A prevalent example of such applications includes on-demand taxi calling services such as Grab and Uber, which have shown to help taxi drivers to reduce the roaming time to search for customers. Regardless of such technology, taxi drivers still have to idly wait until the customers use the application to request taxi services. The ability to accurately predict the customer spawning at a given time could prove crucial in productive taxi roaming. In this paper, we propose a distributed taxi routing optimization approach that aims to suggest each taxi driver to productively roam to areas where the customers are expected to spawn. Specifically, we propose a machine learning model to forecast taxi customer spawning at a given place and time. Then, such a model is used to compute the probability of customer spawning for each road cluster. Finally, a weighted randomization algorithm is used to suggest a taxi driver to travel to a specific region to optimize his search time and to avoid scramble for customers. Our proposed method is validated in the simulation program (COMSET) and compared with the random walk and random destination baselines. The experiment results suggest that our proposed algorithm is effective and yields promising results that overcome both of the baselines.