Teaming up with United Way, a group and I were asked a question: In areas with higher median household income, will there be more
food resources available in that area than areas with lower median income?
To answer this question, our group decided to collect data from the Google Places API and from the Census Bureau. After collecting the data we would create a Tableau page to visulize what we had collected and show our findings to United Way. Lastly, our group wanted to make a machine learning model for the data to test the inital question.
As for the Census Data, we went straight to the source, The American Community Survey. When it came time to put together the Tableau page, our group decided we wanted to make an interactive map of the data. This way United Way both had something to look at and the ability to choose what specifically they might need. At the bottom of the page there are two sets of bar charts. The first one is the sum of resources by census tract, and it accounts for all the data points located in each census tract. The second one is the resources by population and housing units, and it shows the data points based off of each resource per 1000 people, or homes, in a given census tract. Lastly, our group put together a machine learning model to analyze the data we collected and hopefully predict areas that struggle with food deserts. Using Sklearn and TensorFlow, the data was imported and the model was trained. However, it was noticed that the accuracy of the model was inaccurate. This was due to more exturnal factors that play on food deserts than just population size and income levels. Unfortunately the model was unable to predict food deserts the way we wanted, but we where able to analyze the data and map it (as seen in the Tableau section). We concluded that our hypothesis should be rejected because there was not see a strong positive correlation between median household income and access to food resources. Our data indicated that there is actually a slight negative correlation between median household income and food resources meaning, if you live in a census tract with a lower household median income, you are slightly more likely to have access to more food resources than a census tract with a higher household median income.
To try and explain why this occured, we took another look at the data. Uppon further inspection, it appears most of the Topeka, KS area census tracts fall into the lower end of the household median income scale, which is one reason why the food resources might skew towards the lower end. If we were to measure access to food resources against median household income across the entire US, it is possible on average we might find a different result. However, our results do show that population or the density of housing in a given area does seem to have a stronger correlation to access to food resources pandas, censusgeocode, Machine Learing