Saturday, December 18, 2010

Health Analytics Challenge

The Heritage Provider Network is offering a $3 million prize for an algorithm that can predict future hospitalizations based on patient data. This seems like a very cool application of machine learning. If I knew more about machine learning (or anything, really :P), I would totally participate. This sort of application would be invaluable to the health care field. Imagine having the ability to predict future problems with a patient, years in advance. That would cause a huge improvement in the quality of life.

If any of you know anything about machine learning, you should try out this challenge. It's set to start around next year. You could make a huge difference in the quality of life of people. I suppose $3 million would be nice as well. :P


  1. I'm skeptical of the motivations behind this prize. They don't seem too well thought out, or at least not as obviously benevolent as it might immediately appear.

    We'd be foolish to think that we could directly predict anything in a field which we do not yet fully understand. We might give guesses based on indirect data--things like time-occurrences and reported magnitudes of symptoms, but we're still only guessing.

    At worse, such a system will generate self-fulfilling treatments, and at best, we might prevent actual illnesses, but we wouldn't know which actually happened.

    More than anything, I think this is a prize for the actuary and not for the doctor.

  2. One of the main motivating factors of using machine learning is that it can detect patterns in data that humans cannot. Humans aren't capable of analyzing that much data for these patterns. Computers are much better.

    This algorithm would, of course, only generate probabilistic results.

    The point of applying machine learning algorithms to this is so that they can detect patterns that humans can't. This is the real goal of this challenge. Being able to predict hospitalizations is just a ramification of knowing more about health patterns.