1
$\begingroup$

I have collected diabetes patient details. Each user details contains his symptoms, he might have diabetes in symptoms and might not have.

Currently what I am doing:

If patient A have x and y symptoms. And we find 20 patients who have x and y along with diabetes as a symtoms out of 100 patients. So we count A has 20℅(20/100) probability of having diabetes.

Is this correct way of predicting diabetes possibility for the patient?

I appreciate if someone help me to make it more intuitive or may way to play with probablity value.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
3
$\begingroup$

Yes, assuming that the 100 patients are all the patients from your patient pool that exhibit symptoms x and y, including the 20 who do have diabetes.

So, your calculation is a first estimate of the probability that a patient from your pool has diabetes.

Note however, that your pool of patients seems to include only "patients", which are people with symptoms. So your estimate may only be an estimate for "patients". Maybe not for healthy people.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks. But I think most of user who have some health issue they will only use this system. So we can consider all users as patient only, isnt? There is case where user has x,y,z,a,b symptoms and only x and y symtoms matches. Any other thing can be retrived from this? $\endgroup$
    – user123
    Commented Sep 7, 2016 at 11:32
  • $\begingroup$ I think so too, but I'd rather know for sure. If you believe all your "users" are similar to the patients in you pool, then you're fine. Otherwise, you'd have to add a large number of people's details to your data and see by how much your probabilities stay the same. Essentially, you'd be interested in the prevalance of x and y in the larger population if you wish to generalise beyond your patient pool. $\endgroup$
    – Ytsen de Boer
    Commented Sep 7, 2016 at 11:42
  • $\begingroup$ thanks for your input. Can you please review - ec2-52-41-34-62.us-west-2.compute.amazonaws.com:8000 it will give you little clear picture $\endgroup$
    – user123
    Commented Sep 7, 2016 at 11:46
1
$\begingroup$

Beside probabilistic approach, you can use parametric (bayes ) or non-parametric classification models such as decision tree, NN or SVM .

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.