Acta Informatica Malaysia (AIM)

Poverty remains a pressing socio-economic issue in Anambra State, Nigeria, necessitating data-driven strategies for accurate assessment and policy action. This study applies machine learning techniques to model poverty status using socio-economic variables, including age, satisfaction level, perception of poverty trends over the past eight years, choice of health facility, source of fuel, and educational attainment. The analysis utilizes secondary data from the Anambra Bureau of Statistics Poverty Index Survey 2021, comprising approximately 2,500 households across 188 communities. Three classification algorithms: Random Forest (RF), Support Vector Machines (SVM), and Gradient Boosting (GB) were employed to estimate poverty status and compared using key performance metrics: accuracy, precision, recall, F1-score, Area Under the Curve (AUC), Mean Squared Error (MSE), and R-squared. The study’s objectives were to: (1) identify key socio-economic determinants of poverty, (2) apply RF, SVM, and GB models to classify poverty status, and (3) determine the most effective classifier based on predictive performance. Empirical results showed that the Gradient Boosting model had the highest classification accuracy (92.3%), followed by RF (89.7%) and SVM (85.4%). F1-scores ranged from 0.81 to 0.91, with GB outperforming others due to its superior handling of complex, non-linear data patterns. Feature importance analysis revealed that perception of poverty rate and choice of health facility were the most influential predictors, followed by educational qualification and fuel source. These findings demonstrate the value of machine learning in socio-economic research and advocate for its integration into real-time poverty monitoring and targeted policy interventions in Anambra State.