Application of machine learning to investigate heterogeneity in users’ perception of intercity train service quality in developing countries

Abstract

Intercity train service has long been perceived as the most reliable, economic, and safe mode of transportation for long-distance travel in almost all parts of the world, especially the developing countries. Thousands of people from all walks of life use this service every day and therefore, it would be naive to consider the perception of homogenous quality of service across the entire cross-section of users usually anticipated by the service providers to be appropriate. In this study, the effect of heterogeneity in users’ perception of service quality (SQ) on intercity train service was investigated. This case study was performed using a dataset collected in Bangladesh, consisting of responses from 1037 users on eighteen SQ attributes. First, a K-Modes cluster analysis was performed to divide the overall users into a number of smaller but homogenous user groups followed by decision tree analysis with each of these groups. The decision tree analysis helps in the evaluation of how each of the SQ attributes affects the overall evaluation of the SQ by that cluster of users. Three types of trees – Single Tree (ST), Random Forest (RF), and Gradient Boosting Machine (GBM) were generated for each of the clusters generated in cluster analysis. The study showed that either RF or GBM outperformed ST analysis for all six clusters. Rankings of the attributes based on how they were perceived by different groups of users were also evaluated. Female harassment was revealed as the single most important issue in intercity train service. Alongside this, ticket purchasing, on-time performance, noise insulation are some other issues perceived as important by various groups of users.