Who performs better? The heterogeneity of grain production eco-efficiency: Evidence from unsupervised machine learning

This study contributes to the existing literature by providing evidence for the microheterogeneity of agricultural eco-efficiency with machine learning techniques. Using the comprehensive dataset from the "China Rural Revitalization Survey" (CRRS), we employ unsupervised machine learning via the K-means clustering algorithm to dissect the heterogeneity of grain production eco-efficiency from the perspective of farmers. Our findings reveal the classification of grain producers into three distinctive groups: large-scale farmers, conventional self-sufficiency farmers, and novel smallholders. Notably, while large-scale farmers exhibit high grain production volumes, they concurrently generate substantial carbon emissions, reflecting the lowest level of eco-efficiency. Conversely, the novel smallholders emerge as a promising policy inclination due to their superior eco-efficiency, while conventional self-sufficiency farmers exhibit relatively lower eco-efficiency levels. Consequently, we argue that improving grain production eco-efficiency should fully consider the heterogeneity of millions of producers. Overall, this study provides a new perspective that enriches our understanding of the heterogeneity of grain production eco-efficiency, which is crucial for enhancing the effectiveness of policy interventions.