I'm taking a course on AI and machine learning and we've just been recently introduced to k-NN, naive-bayes and 1R. During class I began to wonder if it was always possible to extract a pattern no matter how complex the data was, and what the underlying mathematics that governs this was.
For example, lets say you have a ranking system and ranks are based on various attributes. If there are hundreds of attributes, and each one has an unknown weighting, as well as an unknown contribution to the weighting of OTHER attributes, is it possible to extract any reasonable pattern?
In this particular case, even if we have infinite training data will it be exponential time complexity trying to figure out the how the value of one attribute changes another? Does this make it unfeasible to extract a pattern?
Being new to the field, my terminology is probably wrong so Ill use an example.
If we have a training data it will be of the form:
rank, attribute A, attribute B, C, D, ... (assuming the number of attributes is in the hundreds)
It's known that each attribute does not contribute equally to the rank. It's known that the value of an attribute can possibly change the contribution of OTHER attributes to the final rank. So if attribute A is high for some data, then attribute B, C and D contribute significantly more, but perhaps attribute H, K, W attribute less.
This isn't homework or anything, I've just can't stop thinking about it since class and I didn't want to ask my teacher in case she thought I was an idiot.
[link] [8 comments]