Symptoms to Disease predictor going bezerk

Hi! I’ve been trying to make a simple Symptoms-Disease predictor, which is to be used by real doctors. Since, this seems pretty straightforward, I tried classical approaches: decision trees, naive bayes, etc. but all gave stupid output like correlating somthing like Fever --to–> cancer or debictus ulcer or something, instead of things like a Flu or infection.

Do I need to put some inductive biases in the model or something…?

Let me know what you think here or my email:

Resources I used:
model, notebook attached here: Google Colab) dataset: Disease

Hi Amit

No solutions but perhaps back to basic with the expert systems of the early 80’s.

So for Symptom ! and then 1.1 then 1.1 its improbably 0.7 illness A

So headache → runny noise → sore throat → history of hay fever → pollen time = hay fever
So → no history → cold virus
So → no sore throat - > excessive alcohol -= social consequence

So this is fairly decision tree material but I wonder if each linear layer could be each level of questioning.
Sorry I do not have a solution, Can the symptoms be real world be manifestation which reflect the underlying latent factors because the human body has few tools to react to every external event such as infection, damage or poor performance of the organs or transmission systems: energy, sensor, or endocrine messaging,

Regards Conwyn