EVALUATING AND PREDICTING INTERVAL EFFICIENCIES OF INDIAN IT COMPANIES VIA INTEGRATED NETWORK DEA AND MACHINE LEARNING APPROACH

Abstract

The present paper aims to develop a directional distance function (DDF) based network data envelopment analysis (DEA) approach with internal structure wherein each interrelated process (division) possesses its individual inputs and outputs, linked via intermediate products which are produced and consumed within the system. It effectively handles negative data and undesirable resources with data uncertainty of interval form, and measures the overall system and divisions’ interval efficiencies in pessimistic and optimistic environments. Further, to impart the predictive capability in the proposed approach, it is integrated with least squares support vector regression (LS-SVR) algorithm. This integrated approach not only predicts the performance of newly added DMU in advance that saves the computational time/memory by preventing re-processing of complete network DEA model but also useful in analyzing situations like merger, bankruptcy etc. beforehand. To demonstrate the proposed approach’s practical usefulness, it is applied on 300 Indian Information technology (IT) companies for the years 2018–2021 to estimate system efficiency score(s) in network framework comprising of two divisions, namely, productivity and profitability, and further to predict them using SVR and LS-SVR algorithms. The findings reveal better testing accuracy of LS-SVR algorithm in comparison to SVR based on metrics like mean square error, mean absolute percentage error, etc in both environments. The impact of productivity is more on overall system efficiency in comparison to profitability. Moreover, most of the small-cap and lower-mid-cap companies are found strong and effective relative to upper-mid-cap and large-cap companies.