Predicting Health Outcome from Purchasing History Using Machine Learning

It is well known that life style and dietary habits correlate to certain medical conditions such as Heart Disease, Stroke, and Diabetes. A type of big data that reflects people’s life style and dietary habits is their purchasing history. People who are prone to certain medical conditions might show preference towards certain food types. Purchasing history of a simulated population of 100,000 individuals is used to demonstrate that buying people’s purchasing history records from big companies could be a worthwhile investment by the public health agencies. A Neural Network that uses the Kohonen’s Self Organizing Feature Map (SOFM) is used on the purchasing data to group people into categories based on their buying habits. Individuals who have already been diagnosed with a medical condition can be identified by their purchase of certain prescribed medications. The segment of the population that clusters into a group that includes these individuals is predicted to be at risk for the same medical condition. Numerical results supporting the Big Data analytic design and validation are also presented.