.png){kind=icon}
.png){kind=icon}
Return
Hybrid DEA-machine learning framework for predicting multiple targets
.png){kind=icon}
PRE
AI
DOI：10.1080/03155986.2026.2630138.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study proposes a novel hybrid framework that combines data envelopment analysis (DEA) with multi-target machine learning (MTML) to evaluate and predict the performance of new decision-making units (DMUs) in large-scale datasets. Although prior research has integrated DEA with ML, most studies have been confined to predicting a single efficiency score and have not provided benchmarking information such as reference sets and improvement targets which is central to DEA's practical value. These limitations reduce the applicability of such models for realistic decision support, because inefficient DMUs require not only efficiency scores but also feasible and balanced improvement targets. Moreover, conventional approaches often necessitate rerunning the DEA model entirely whenever new DMUs appear, imposing substantial computational costs in the big-data era. To address these gaps, this study introduces a multi-target prediction approach capable of simultaneously estimating CRS and VRS efficiency scores, reference sets and return-to-scale (RTS) information for new DMUs. We examine three multi-target modeling strategies: (1) a DEA-coupled multi-target to single-target transformation, (2) direct multi-target modeling (DMM) and (3) multi-task learning (MTL). By expanding DEA-based prediction from a single indicator to multiple outputs, the study offers both theoretical contributions and practical value for benchmarking applications.
Keywords:
Data envelopment analysis
multi-target machine learning
efficiency predictions
benchmarking
large-scale DMU dataset