Passengers' requests clustering with k-prototype algorithm for the first-mile and last-mile (FMLM) shared-ride taxi service

Abstract

Shared mobility services are evolving globally. However, the first-mile and last-mile (FMLM) shared-ride taxi service poses a complex problem due to its large-scale nature and mixed-type variables (numeric and categorical features). As the input size of the problem increases exponentially, the absence of a known polynomial-time algorithm further complicates the finding of an optimal solution. Consequently, exploring potential solutions becomes computationally infeasible for more significant instances. Thus, this paper proposes using the k-prototype algorithm, an unsupervised learning approach, to cluster passengers' requests for FMLM shared-ride taxi service, which can reduce the problem's complexity via feasible clustering. Notably, the k-prototype algorithm is suitable for data sets with both numeric and categorical variables. It demonstrates a promising ability to handle large data sets effectively. As presented in this paper, the FMLM shared-ride taxi service prototypes and their unique characteristics could be optimally identified using the k-prototype algorithm with the Silhouette coefficient (as a performance index). By examining an illustrative case study with ten mixed-type variables of 946 passengers' requests, the results demonstrate the effective clustering of passengers' requests into three distinct prototypes, which can be characterized uniquely based on the temporal factors (pickup time of individual requests) and trip characteristics (including traveled distance, taxi type, as well as pickup and drop-off locations) that are significant in operating a competitive shared-ride taxi service. This paper is anticipated to reveal useful practical implications for the relevant stakeholders, especially the taxi service providers, in managing the FMLM shared-ride taxi services optimally to ensure an efficient and effective operating system.