Comment by teiferer
What is an "accuracy" of 83%? Do 83% of predicted diabetes cases actually have diabetes? Or did 83% of those who have diabetes get diagnosed as such? It's about precision vs. recall. You can improve one by sacrificing the other. Boiling it down to one number is hard.
They use the area under the receiver operating curve, which is a pretty standard way to boil that down to one number.