DETERMINING PREDICTORS OF POVERTY AT GAWAD KALINGA-PIELA COMMUNITY, CITY OF DASMARINAS, PHILIPPINES USING MACHINE LEARNING

Abstract

This study utilized a comprehensive dataset containing demographic, economic, and health-related variables from a community survey to identify significant predictors of poverty incidence among households. The analysis involved preprocessing steps such as missing value imputation, categorical variable encoding, and irrelevant feature removal. Dimensionality reduction was performed using Principal Component Analysis (PCA) to retain 95% of the dataset's variance, simplifying the feature space for subsequent modeling. Logistic Regression, Random Forest, and Support Vector Machine (SVM) models were evaluated, with Logistic Regression further refined via Grid Search CV to optimize regularization strength and penalty type. The best-performing Logistic Regression model achieved an accuracy of approximately 71.43% and an ROC-AUC of 64.44%. Key components influencing poverty predictions were traced back to original features, highlighting the roles of occupational types, health practices, disaster risk reduction, community support, and educational opportunities. These findings provide valuable insights for policymakers and community planners aiming to mitigate poverty, demonstrating the impact of socioeconomic factors, health, and education on poverty levels