A data envelopment analysis-informed neural network for production frontier estimation with soft regularization of production axioms

Article Info

Authors: Xingyue Chen, Min Yang, Zixuan Wang, Liang Liang
Journal: Computers & Operations Research
Publication: December 2026, Volume 196, 107636
DOI: 10.1016/j.cor.2026.107636

Overview

The article introduces the Data Envelopment Analysis-Informed Neural Network (DEAINN), a model designed to estimate production frontiers while retaining the theoretical grounding of Data Envelopment Analysis (DEA). It responds to DEA’s overfitting and computational limitations and to the theoretical weaknesses of purely data-driven neural networks.

Proposed Method

DEAINN uses a multi-layer feed-forward artificial neural network with a DEA-informed loss function. Soft regularization incorporates monotonicity, concavity, and the envelopment property into training. The model penalizes violations of these production axioms, guiding the network toward economically consistent frontier estimates without imposing a large set of hard constraints.

Evidence and Application

Monte Carlo simulations compare DEAINN with DEA, Corrected Concave Non-parametric Least Squares, DEA-ANN, and Least-Squares Boosting–Multivariate Adaptive Frontier Splines. The article reports competitive estimation error, favorable generalization, and lower computation time as sample size increases. An empirical application evaluates energy utilization efficiency across 282 Chinese cities.

Key Takeaways

• DEA assumptions can serve as soft regularization for neural-network frontier estimation.
• The approach targets overfitting, scalability, and out-of-sample generalization.
• The reported city-level application demonstrates practical energy-efficiency use.

Cite this paper

APA: Chen, X., Yang, M., Wang, Z., & Liang, L. (2026). A data envelopment analysis-informed neural network for production frontier estimation with soft regularization of production axioms. Computers & Operations Research, 196, 107636. https://doi.org/10.1016/j.cor.2026.107636

MLA: Chen, Xingyue, et al. “A data envelopment analysis-informed neural network for production frontier estimation with soft regularization of production axioms.” Computers & Operations Research, vol. 196, 2026, article 107636. https://doi.org/10.1016/j.cor.2026.107636.

IEEE: X. Chen, M. Yang, Z. Wang, and L. Liang, “A data envelopment analysis-informed neural network for production frontier estimation with soft regularization of production axioms,” Computers & Operations Research, vol. 196, Art. no. 107636, 2026, doi: 10.1016/j.cor.2026.107636.

Source

Read the original article at ScienceDirect for complete source details.

View the ScienceDirect article