Best output prediction in OECD railways using DEA in conjunction with machine learning algorithms

Abstract

Efficiency measurement plays an increasingly important role in the regulation and management of railway organizations. Despite its proven usefulness in efficiency measurement, data envelopment analysis (DEA) lacks predictive capability. In order to benefit from their learning and mapping capabilities, machine learning (ML) algorithms have been used as a complementary method to DEA, recently. However, the majority of the existing ML-DEA studies focused on efficiency estimation while disregarding the prediction of DEA projected inputs/outputs toward better performance. This study proposes a novel framework using the adaptive neuro-fuzzy inference system (ANFIS) and the support vector machines (SVM) models in conjunction with the context-dependent DEA model to predict efficiency scores and the best input/output levels for 37 railway companies of OECD countries. Despite drawing on a small sample size, the proposed DEA-ANFIS and DEA-SVM models successfully predicted the efficiency scores and the best output levels of the organizations via approximating the efficient frontiers.