mgeracie/malawi_hunger

We create a local countrywide estimation of the burden of hunger in Malawi with local confidence intervals. # Estimating the Geographic Distribution of Hunger in Malawi **Contents:** * Background * Methods * Results ## Background In this project we create a country-wide local estimation of the burden of hunger in the country of Malawi and evaluate the systematic error and statistical uncertainties of our model. The Malawian economy is one of the least developed in the world and faces persistent problems due to poor infrastructure and government corruption. Our goal is to create a model that will be useful for government and supporting organizations to address areas with the greatest need. ## Methods The burden of hunger is measured by the expected number of days/week that the average household at a given location is required to restrict meal portions. Our model is trained on survey data from over 12,000 households in 2016, along with their geographic coordinates, and is generalized to the rest of the country on the basis of local poverty levels, population density, and market access. The geographic data was downloaded from the RCMRD Geoportal, a platform which provides open source geospatial maps for eastern and southern Africa. They are stored as raster data which we read into numpy arrays. The survey data is taken from the Integrated Household Panel Survey of Malawi performed in 2016 for the Living Standards Measurement Study (LSMS). A histogram of survey responses can be seen below. We join this data using the geographic coordinates of the survey households and train a variety of machine learning models, including decision tree regressors, linear regressors, k-nearest neighbor regressors, and support vector machines to predict the days/week of portion restriction from the independent variables. Finally, we feed these models into a stacking regressor which lowers bias and reduced the chance of over-fitting. This done, we generalize the survey data to a local country-wide estimation by querying the value of the independent variables at each point in the cou …