Modelling Livelihood Security of Tribal Farmers in South Odisha using Machine Learning

Tribal farming systems ensure livelihood security through complex socio-economic and behavioural interactions that defy simple linear models. The study analysed primary data collected through simple random sampling method from 180 households in Gajapati and Rayagada districts of Odisha during 2023-24 to analyse the Livelihood Security score using a Random Forest regression. Out-of-bag validation demonstrated model stability with an R² of approximately 0.865 using around 400 trees. The age was the most significant predictor, followed by self-confidence, with smaller contributions from management orientation and innovative proneness. One- and two-dimensional partial dependence outcomes highlighted non-linear age effects and interactions, indicating that increased confidence and enhanced management capacity improve predicted livelihood security across all age groups. These results suggest actionable strategies for agricultural extension: implementing confidence-building and management training tailored to life-stage constraints could yield substantial benefits. Limitations include the correlational nature of the data and the reliance on partial dependence.