HN Explorer

Ask HN: what are simple problems where Machine Learning can help?

8 points by toisanji ↗ HN
I've been doing a lot of machine learning lately, applying models to do classification and regression. I'm trying to find simple problems that people have where machine learning could help. I would love to have a discussion about it.

7 comments

[ 3.2 ms ] story [ 28.9 ms ] thread
VorticonCmdr
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.
toisanji
To clarify, I was thinking more along the lines of startup ideas.
tptacek
Isn't that a little like asking about simple problems that graph theory or compiler theory could solve as the basis of a startup?
toisanji
True, but sometimes you do come across some interesting ideas.
cmaxwe
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.

sharemywin
-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