I got an email recently from John Toczek, an OR professional with Aramark who writes the PuzzlOR column for OR/MS Today):
You ran an article on a contest I created back in 2010 called the Analytics X Competition [http://mat.tepper.cmu.edu/blog/?p=1024] where contestants were trying to predict where homicides would occur in Philadelphia zip codes. That contest ended some time ago but I am working on two new projects that I thought might interest you.
The first is a medical diagnostic site (www.whatsmydiagnosis.com) that uses the correlations between symptoms and diseases to predict diagnoses. The site ranks the likelihood of all diseases in descending order based on the symptoms you select. It’s in the proof of concept phase right now but I wanted to get the word out in order to get some feedback from the community so I can improve it.
The second is an analytics treasure hunt that is due to run in Analytics magazine on July 9th. (www.puzzlor.com then click on “The 2012 Analytics Treasure Hunt” link.) It’s basically a series of 5 puzzles that when completed form a latitude and longitude where I have hidden $100. If you’ve ever heard of GeoCaching, it is similar to this. My hope with this project was to get the word out about O.R. and hopefully attract new people to the field.
I checked out the medical diagnosis site. After putting in my sex (Male), height (5’10” and above) and weight (over 180 lb: seriously? in a land of morbese obesity, over 180 is the maximum grouping?) and found …. a 42.5% likelihood of depression. That’s a bit of a downer. In fact, I’m feeling kinda depressed about the whole thing. Hey! It works!
I wouldn’t take the site too seriously (as the About page there says: should not be used as professional medical advice) but perhaps as more data gets entered, the predictions will get better.
As for the treasure hunt, I am looking forward to the challenges (available July 9). I hope the hunt is difficult, though. I am currently in Vilnius, Lithuania and will be for most of the week. Either I luck out and John has decided that Lithuania is the ideal place to hide his $100 or it will take a while for me to get to where the money is hidden. That’s assuming I can solve his challenges, of course.
Idea of diagnosis site is pretty interesting.
As for treasure hunt, assuming all academics need such thrills, it is even more interesting. Even the name of the puzzle attracting us for at least having a look at it. We need to use this one week wisely – until you arrive!
If it’s any consolation, I checked “stiff joints” and “muscle cramps” and was told there’s an 89.1% chance I’m diabetic. Could be worse, I suppose — could be 89.2%.
I entered Male, Height 5’5″-5’9″, Age 41-60, Weight 181+, and as a sole symptom Back Pain. The site tells me I have Kidney Failure with “Likelihood” 154.8%!! Wow — that’s a confident diagnosis. Happy to report that my kidneys are fine, as is my back — I was exploring to learn the marginal effects of symptoms one at a time on the likelihood of different diagnoses.
By contrast, I put in both back pain and stiff joints (on separate attempts) and was told I had the flu both times.
Yes, the diagnosis idea is cool. But it’s not at all new. Some 40 years ago, at Stanford, some CS people developed Mycin, which was arguably the first medical expert system ever.
BTW, here’s an interesting paper:
Categorical and Probabilistic Reasoning in Medical Diagnosis (1978) [pdf]