Interesting idea -- I have a few questions about how this would work:
1. What's your dependent variable? If it's a dichotomous outcome (acceptance/rejection), you'd need something more sophisticated than run-of-the-mill regression -- particularly if you're interested in calculating predicted probabilities.
2. How do you operationalize your independent variables? This would be particularly consequential in the case of things such as "fit" or "research experience," which an applicant may be unable to judge objectively.
3. There are likely many latent (i.e. unobserved) confounders that influence a given school's admissions decisions; for example, subfield quotas, applicants' adherence to the dominant theoretical orientations within a subfield, faculty availability, etc. How would your model account for these factors?
Again, very interesting idea -- however, you should be sure that your model is accurate before soliciting others' information online.