Diabetes Prediction
With the data refinery hitting its stride, we wondered if lab results could be used to predict health conditions through technology only. I was convinced we could. With the help of a small team, we produced a case study to predict diabetes development using a narrow data set excluding glucose test results. The model was correct with an 87% accuracy. The results also highlighted a surprise: a calcium in urine test as a highly relevant variable. Doctors that confirmed decreased kidney function occurs with diabetes, so leaching calcium into the urine made sense and explained the models non-obvious conclusion!