EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

1777. Adaptive Constrained Enveloping Splines and Random Forest for Technical Efficiency Measurement

Invited abstract in session MA-48: DEA and its application, stream Data Envelopment Analysis and its Application.

Monday, 8:30-10:00
Room: 60 (building: 324)

Authors (first author is the speaker)

1. Víctor Javier España
Center of Operations Research (CIO), Miguel Hernández University of Elche
2. Juan Aparicio
Center of Operations Research, Miguel Hernandez University of Elche
3. Josep Xavier Barber
Center of Operations Research (CIO), Miguel Hernández University of Elche

Abstract

In various research fields, understanding the relationship between predictors and a response variable involves curve estimation with specific characteristics. For instance, isotonic regression estimates mortality rates, assuming an increasing relationship between older age groups and mortality risk. Similarly, efficiency analysis can be reframed as a shape-restricted regression problem, aiming to estimate a non-decreasing and concave function that envelopes the observed data points. In this context, Data Envelopment Analysis (DEA) is commonly used for nonparametric production frontier estimation. However, DEA is susceptible to overfitting, resulting in overly optimistic efficiency estimates.

Recently, an adaptation of the Multivariate Adaptive Regression Splines (MARS) algorithm was introduced for production function estimation, addressing overfitting concerns. Our work builds upon this methodology, with three primary objectives. First, we propose a method to incorporate variable interaction during model fitting while maintaining shape constraints for production functions, enhancing predictive capacity. Second, we enhance robustness by randomizing data and input variables during model construction, drawing inspiration from the Random Forest (RF) methodology. Finally, within the RF framework, we can identify the most relevant inputs related to output prediction.

Keywords

Status: accepted

Back to the list of papers