Profit efficiency estimation and prediction of banks with mixed structure: a unified DDF-based network DEA and ML approach

The financial health of a country is significantly influenced by the profitability of its banks, which arises from their multiple inter-related core operations. Effective strategic planning requires precise estimation and prediction of the performance for each operation. The present study designed a mixed-network structure of banking operations that integrates operational reserves and diversified financial services, including fund deployment and digital/non-digital services. A directional distance function -based network data envelopment analysis (DEA) model is proposed to comprehensively evaluate the profit efficiency of individual divisions and the overall system, considering competitive scenarios with varying input costs and output prices, undesirable outputs, and handling negative data. Further, value-based input–output targets are derived for providing improvement path to inefficient units. To address the computational complexities of network DEA, machine learning techniques namely, support vector regression and artificial neural network have been merged with the proposed framework to forecast the profit efficiency of newly added banks. This methodology has been applied to the Indian banking sector over fiscal years 2013-2023, using a five-year DEA window analysis. The findings show a consistent upward trend in profit efficiency across all divisions, with banks improving their capabilities in fund generation, deployment, and digital/non-digital services. Additionally, the integrated approach yields highly accurate profit efficiency predictions evaluated for window $$\text {W}_7$$ W 7 (2019–2023). Moreover, as evidenced by metrics like mean square error ANN has outperformed SVR in performance and accuracy.