Decoding the future of agricultural participation: machine learning insights to unravel the plausible triggers

Introduction As agricultural participation continues to shift under the pressures of urbanization, climate change, and evolving socioeconomic conditions, understanding the drivers behind household engagement becomes increasingly vital. Methods This study explores these dynamics using household budget survey data, applying decision trees, random forests, and gradient boosting to uncover trends in model performance and variable importance over time. Results Our comparative analysis reveals a consistent decline in decision tree accuracy, which reflects the model's limited ability to capture increasingly complex and non-linear relationships in household behavior. In contrast, ensemble learners—random forests and gradient boosting—combine multiple weak learners, typically shallow decision trees, to improve predictive performance. Random forests aggregate predictions through bagging, while gradient boosting builds trees sequentially to correct prior errors. These methods demonstrated superior sensitivity and balanced accuracy in identifying agricultural participants, particularly by 2017–2018, when random forests achieved a notably low out-of-bag error rate for classifying agricultural sales participants. However, early-year specificity remained a challenge. Key predictors evolved from income-dominated variables in 2002–2003 to a more nuanced mix of household size, age, water access, and geographic context by 2017–2018. While all models identified overlapping predictors, ensemble methods were more effective in capturing subtle interactions and demographic shifts. Decision trees, though less accurate overall, provided valuable insights into spatial variation, especially in 2010-2011 when district-level factors were prominent. Rural households consistently showed higher participation rates, with urbanization and regional disparities becoming increasingly influential. Discussion These findings highlight the strength of ensemble learning in capturing the complexity of agricultural engagement and underscore the need for adaptive, data-driven policy strategies. The observed shifts in variable importance reflect a changing socioeconomic landscape, calling for targeted interventions that address local realities and emerging challenges such as climate volatility and rural-to-urban migration.