Structured neural network model reveals how resettler perceptions influence agricultural effectiveness

Abstract Large-scale dam projects often lead to displacement, posing serious challenges to agricultural livelihood reconstruction. This study explores how subjective perceptions influence agricultural production effectiveness among resettled households in Guizhou Province, China. While traditional models assume linear responses, such assumptions may overlook how displaced farmers adapt to new environments. We propose a hybrid SEM-BP model integrating Structural Equation Modelling with a neural network. This approach is designed to capture complex, non-linear behavioral relationships often missed by traditional methods. Using survey data from 832 households, the SEM-BP model outperforms standard SEM in explanatory power. Results show that perceptions of production security and environmental suitability significantly affect agricultural effectiveness, with agricultural production intention acting as a mediator. In contrast, perceptions of production structure and policy proved insignificant. This result highlights that resettlers prioritize immediate risks over broader structural or policy factors. These findings underscore the importance of incorporating non-linear behavioral factors into analyses of livelihood adaptation. The SEM-BP approach offers practical insights for designing policies that better support displaced farming communities.