Integrating Bootstrap DEA and Machine Learning Meta-Frontiers to Assess Technical Efficiency and Technology Gaps in Gherkin, Paddy, and Groundnut Farming Systems in Northern Tamil Nadu

Abstract

This paper analyses technical efficiency and technology gaps among 305 smallholder farmers cultivating gherkin, paddy, and groundnut in northern Tamil Nadu using a dual-frontier framework. Crop-specific technical efficiency is estimated via input-oriented Data Envelopment Analysis under variable returns to scale with bootstrap bias correction. To account for technology heterogeneity across crops, a DEA meta-frontier is constructed and Technology Gap Ratios are derived. In parallel, Quantile Random Forests are employed to estimate flexible nonparametric production frontiers at the 95th conditional quantile, and a pooled machine-learning meta-frontier is obtained. Results indicate high technical efficiency across all crops, with paddy and groundnut operating closer to their group frontiers, while gherkin exhibits greater dispersion. Meta-frontier estimates show minimal technology gaps, with gherkin technology lying closest to the global frontier. Strong agreement between DEA- and ML-based frontiers confirms the robustness of the findings.