Using Random Forest to Classify Poverty Levels in Oyo State Communities

Accurately classifying community poverty levels from easily observable variables supports more efficient targeting of social protection programmes in Oyo State, South West Nigeria. This study applied a random forest classifier to predict poverty level classification for communities in Oyo State using household survey data from the 2018 and 2021 Nigeria Living Standards Survey. Data for 3,200 households across Ibadan North, Ogbomoso, and Saki LGAs were used, including variables for education, housing quality, access to services, asset ownership, employment type, and food security. Five poverty level classes were defined according to NBS poverty lines. Random forest achieved 74.6 percent accuracy and a macro-averaged F1-score of 0.71, outperforming ordinal logistic regression. The most important features were housing type, access to electricity, and educational attainment of the household head. Rural households in Saki LGA showed the deepest poverty concentration. Classification accuracy was highest for extreme poverty and non-poor categories, with misclassification most frequent at moderate poverty boundaries. SHAP visualisations provided interpretable explanations of community-level classification drivers. The study demonstrates that random forests can provide interpretable and reasonably accurate poverty classification from standard survey data. Recommendations include deploying the classifier to support Oyo State's conditional cash transfer targeting, regular model retraining with updated survey data, and expansion to cover all 33 LGAs using geographically stratified sampling.

Keywords: random forest, poverty classification, Oyo State, social protection, household survey analytics