Diabetes is a serious disease affecting millions of people across the entire world. Thus, correct and timely prediction of this disease, is very important due to the complications it can have in the case of other life-threatening diseases. Diabetes can be diagnosed through observation of some clinical data, in particular blood glucose level. But if blood glucose level is used to diagnose diabetes, there are other factors (called risk factors) that can suggest, but not confirm, that a patient has diabetes. The latter consideration leads to the development of this project, where we analyze different clinical data, to understand the impact of certain risk factors, on the likelihood of developing diabetes in patients.
In the medical field it is hard to work purely with neural networks, since usually what is required is not a single prediction (classification/regression task) but probabilities associated with each single possible outcome. Of course, this could be obtained with neural networks as well, but with them it is impossible to set up all the causal connections between the features. For such reason bayesian networks have been useful and powerful in extrapolating the values of correlations between risk factors and diabetes in our analysis. The results obtained are in line with the basic knowledge of internal medicine, confirming the validity of the produced work.
This project was done in collaboration with @LucaZucchini97 (https://github.com/LucaZucchini97)