Machine Learning Approach for Decoding the Efficiency of the World's Largest Welfare Scheme ‘MGNREGA’: A Deep Dive Into Welfare State Dynamics and Local Heterogeneity

Abstract

Against the backdrop of liberal capitalist democracies, the welfare state has emerged as a crucial institution that aims to reconcile market-driven economic growth with social justice imperatives. The introduction of MGNREGA in 2005 exemplifies welfare intervention with the objectives of poverty alleviation, livelihood security and the empowerment of marginalised communities. Drawing on concepts such as decommodification and the state-in-society approach, this study evaluates the efficiency of the MGNREGA across India's diverse sociocultural landscape. By employing a combined approach that integrates data envelopment analysis (DEA) and machine learning (ML) models, such as random forest, this study reveals efficient decision-making units. This study identifies key efficiency drivers for exploring the relationships among resource allocation, workforce participation and various development outcomes achieved through MGNREGA across all 726 districts of India. By weaving together the threads of regional disparities, state capabilities and implementation outcomes, this study sheds light on the enigma of MGNREGA's uneven success across diverse states in India. Notably, nationally, 36.09% of the districts are efficient, 22.59% are moderately efficient, and 41.32% are inefficient. In essence, while financial factors are necessary for MGNREGA implementation, socio-economic factors, rural infrastructure development and targeted interventions play pivotal roles in determining efficiency.