Geospatial analysis for mapping poverty in the Philippines

This thesis explores the use of satellite imagery, geospatial data, spatial econometric methods, and machine learning techniques to map poverty in the Philippines for the years 2012, 2015, and 2018. By combining remote sensing data with municipal-level poverty estimates, the study evaluates various models effectiveness in predicting poverty. Spatial econometric models, including the spatial lag model (SLM) and spatial error model (SEM), reveal significant spatial patterns and consistently outperform the ordinary least squares (OLS) model. In addition, the research assesses machine learning algorithms such as generalized least squares (GLS) regression, neural networks, random forest, and support vector regression (SVR). The random forest model emerges as the most reliable predictor and the highest accuracy. Key predictors of poverty include nighttime light, road density, points of interest, and urban land surface, with environmental factors like water availability and drought conditions also playing significant roles. Integrating spatial analysis with advanced machine learning provides a robust framework for understanding and predicting poverty. This approach enhances the accuracy of poverty predictions and guides the development of effective poverty reduction strategies in the Philippines.