eXplainable DEA approach for evaluating performance of public transport origin-destination pairs

Abstract

Understanding public transportation efficiency is crucial to enhancing urban mobility and economic growth. This study aims to evaluate the efficiency of the public transportation system using an explainable data envelopment analysis approach. Specifically, data envelopment analysis and explainable artificial intelligence techniques were incorporated to uncover the black-box relationship between the inputs and outputs of the efficiency score. The efficiency of the public transportation system was defined as travel times for subway, bus, and multimodal modes relative to the number of trips and travel distances. The efficiency score was evaluated by the origin-destination pair unit. As a result, the efficiency score was estimated to be 0.69 on average, indicating that a reduction of 31% in travel times is required to achieve a perfect score of 1.00. Among the 33,313 origin-destination pairs, 39 had a perfect score of 1.00. The results of the interpretation model showed the order of importance for features—buses, subways, and multimodal modes—with SHapley Additive Explanations values of 0.047, 0.029, and 0.022, respectively. These results suggest that focusing primarily on reducing bus travel times is effective for improving overall efficiency. In this manner, the explainable data envelopment analysis helps measure performance, understand results, and suggest improvement directions.