Farm efficiency estimation using a hybrid approach of machine-learning and data envelopment analysis: Evidence from rural eastern India

Abstract

In agricultural sector, farm efficiency evaluation is an important means of farm management. To evaluate the farm efficiency for effective allocation of agricultural resources, data envelopment analysis (DEA) is used. Commonly, a two-stage regression analysis is used to treat the obtained efficiency values on a set of explanatory variables. However, majority of the past studies explained the variables influencing efficiency rather than efficiency prediction. This paper aims to use DEA in combination with Machine learning approach to examine and predict the impact of environmental variables on farms’ performance. Random forest (RF) algorithm has been employed, which is one of the most useful machine learning algorithms of recent times. First, DEA was used to evaluate efficiency of all the farms. Then the RF method was employed to examine the variables crucial in predicting farm performance. This multi-stage model was applied to 450 paddy producers in rural Eastern India. The results of the RF algorithm revealed that land ownership, Kisan Credit Card (KCC), and educational status were the most crucial variables which affected the performance of the paddy producers. With the identification of the major factors influencing agricultural production, new policy actions may be developed to assist the small farmers. Furthermore, this joint DEA-Machine-learning approach may help future researchers not only to investigate but also to predict the impact of the important environmental variables on farms’ performance.