-Facial, Finger print,iris Recognition -detect pulse from images possibly other vitals -game AI -Sales forcasting -recommendation engines -underwriting -summarization article -image, video, text classification

I tried to use NHL box scores to predict NHL outcomes. Seemed like a pretty good opportunity to use machine learning. Ended up only being able to predict at like 54% or something like that which isn't even enough to break even.

What I did find though was that if I looked for games where my estimated accuracy rate was high and the odds were good then I probably could have bet on those games for a profit but they were very rare.

Here are class projects for Machine Learning at Stanford: http://cs229.stanford.edu/projects2012.html

Should be a good start.

spell correction is somewhat of the most popular and easy to understand problems where machine learning can help. For example: several user enter a misspelled word and then correct them in the next step. If this happens often enough you might infer that the second word is most probably the correct version, especially if the second word has a small levenshtein -distance to the first.

To clarify, I was thinking more along the lines of startup ideas.

Isn't that a little like asking about simple problems that graph theory or compiler theory could solve as the basis of a startup?

True, but sometimes you do come across some interesting ideas.