Blog 6

I still remember that I felt half nervous half excited while waiting for the “predict” function to finish, and I felt thrilled while seeing the map with the predicted values. Although we still have to add some more variables to improve the accuracy of the data and population description map, this first step — the template making process— is crucial. After discussing with Professor Frazier, we plan to go further to explore some Mathematical model first and use the sample data from DHS website to create a more continuous and detailed population datasets. One of the most promising method is Hierarchical Bayesian Model (HBM) for this use. We figure out a general outline of our later work. Since we have already had all the layers and run the data with random forest to have clan population prediction. Now we are going to collect some more specific household data from DHS, like the household size and particular locations of them, and put the data into HBM which may involve some multivariable logistic regression models. These kinds of models will have some input which are discrete point data and generate some output which are predicted continuous data. This step will be noteworthy for this whole research. Also, for our research, we are not going to use individual data from DHS but household data. This is a noteworthy part of our study. Some of the other studies before have investigated individual information and map that onto the map. However, there are many aspects of life which are much more strongly related to factors of household. For example, the places where people choose to live is highly depended on where the labor of the family work. Household, or in other words, family is highly important for mapping population and population description. Only several research take household size and related information into account so that our research will be a breakthrough. However, there are also some technical issues waiting to be addressed and most of them will require some more advanced mathematical models. These are mostly what we talked about and are researching on during this period of time. We are studying more and more about the promising method HBM and keep encountering new problems.