A Neural Network approach for Non-parametric Performance Assessment

Abstract

Data Envelopment Analysis (DEA) is the most popular non-parametric method for the efficiency assessment of homogeneous units. In this paper, we develop a novel approach, which integrates DEA with Artificial Neural Networks (ANNs) to accelerate the evaluation process and reduce the computational burden. We employ ANNs to estimate the efficiency scores of the milestone DEA models. The relative nature of DEA is considered in our approach by assuring that the DMUs used for training the ANNs are first evaluated against the efficient set. The ANNs employed in our approach estimate accurately the DEA efficiency scores. We validate our approach by conducting a series of experiments based on different data generation processes and number of inputs and outputs. Also, these estimated efficiency scores satisfy the properties of the fundamental DEA models. Thus, our approach can be employed for large scale assessments where the traditional DEA methods are rendered impractical.