Comparing support vector regression and statistical linear regression for predicting poverty incidence in vietnam

Urban and rural poverty are key issues of the Millennium Development Goals and much research is done on how to reduce poverty sustainable and long-ranging. However, small scale poverty maps at full spatial and temporal coverage are fundamentally necessary but rare. Some small scale poverty mapping methods have been developed in past years, but these methods often rely on data which has to be collected in resource intensive field work. We therefore compare two statistical data mining tools, Support Vector Regression and Linear Regression, to scale Vietnamese poverty data from a coarser training to smaller scaled testing set. The Support Vector Regression performed worse than the Linear Regression model with feature subset. However, the Support Vector Regression model showed a more systematic error which might be corrected more easily than the error of the Linear Regression approach. Furthermore, both models showed dependency on spatial effects. Hence, integration of spatial information might increase the success of future models and turn data mining approaches into valuable tools for poverty mapping on small scales.