A Hybrid Two-Stage DEA and Deep Learning Framework for Efficiency Evaluation of Iranian Stock Exchange Companies

Abstract

This study investigates the efficiency of Iranian stock exchange-listed companies (2007–2023) using a hybrid approach integrating two-stage Data Envelopment Analysis (DEA) with Deep Learning (DL). Traditional DEA evaluates efficiency but struggles with nonlinear patterns and noisy data. By combining DEA with Long Short-Term Memory (LSTM) and TabNet models, this research addresses these limitations. Results reveal that LSTM outperforms TabNet in predicting efficiency scores (MSE: 0.0025 vs. 0.0203), demonstrating its superiority in capturing temporal dependencies in financial data. The hybrid framework enhances accuracy in identifying inefficiencies, optimizing resource allocation, and informing strategic decisions. This methodology bridges DEA’s multi-input/output assessment with Artificial Intelligence (AI)’s predictive power, offering transformative insights for financial analytics.