Profiling Socially-Structured Vanpooling Users in Oman: A Data-Driven Approach

Abstract

Understanding the social preferences of ridesharing users is very important for the implementation of user-oriented transportation services in several countries throughout the Middle East. This paper reports preliminary results on profiling socially-structured vanpooling users in Oman. We adopted a three-step "cluster-then-classify" approach to analyze a dataset of 3,615 current and potential ridesharing users that we collected from the various regions of Oman. In the first step, we applied the K-Modes clustering (an unsupervised labeling algorithm of categorical data) to identify distinct clusters of riders. Five profiles of riders were identified: "reserved students", "broad-minded students", "independent workers", "dependent workers", and the "unemployed". In the second step, we compared the performance of four algorithms (Decision Tree, Random Forest, CatBoost, and Logistic Regression) to classify riders into the five identified classes, leading to an accuracy of 0.91. In the third step, we performed an additional test using a new dataset, and compared the performance of the four classifiers against the modes of each cluster. A best similarity percentage of 90.24% was obtained, suggesting that the five clusters satisfactorily partition riders into distinguishable and interpretable classes. To the best of our knowledge, this is the first research work that applies such an approach to profile shared mobility users in the MENA (Middle East and North Africa) region. The obtained results are expected to help transportation services better address the preferences and needs of riders in Oman and other MENA countries with similar socio-cultural contexts. Abstract Understanding the social preferences of ridesharing users is very important for the implementation of user-oriented transportation services in several countries throughout the Middle East. This paper reports preliminary results on profiling socially-structured vanpooling users in Oman. We adopted a three-step "cluster-then-classify" approach to analyze a dataset of 3,615 current and potential ridesharing users that we collected from the various regions of Oman. In the first step, we applied the K-Modes clustering (an unsupervised labeling algorithm of categorical data) to identify distinct clusters of riders. Five profiles of riders were identified: "reserved students", "broad-minded students", "independent workers", "dependent workers", and the "unemployed". In the second step, we compared the performance of four algorithms (Decision Tree, Random Forest, CatBoost, and Logistic Regression) to classify riders into the five identified classes, leading to an accuracy of 0.91. In the third step, we performed an additional test using a new dataset, and compared the performance of the four classifiers against the modes of each cluster. A best similarity percentage of 90.24% was obtained, suggesting that the five clusters satisfactorily partition riders into distinguishable and interpretable classes. To the best of our knowledge, this is the first research work that applies such an approach to profile shared mobility users in the MENA (Middle East and North Africa) region. The obtained results are expected to help transportation services better address the preferences and needs of riders in Oman and other MENA countries with similar socio-cultural contexts.