Jacob J. Walker's Blog

Scholarly Thoughts, Research, and Journalism for Informal Peer Review

Non-Linear “Regression” using a Pattern Recognition Algorithm combined with Monte Carlo Simulations

without comments

For a bit of time, I have been trying to find a single standardized method of data mining to test any set of data across various curves of best fits with various underlying error distributions.  (As I have realized my idea of removing “outliers” isn’t really solving the problem, because they are often there for a “reason”) Yesterday, I had a “crazy” idea that might just be able to solve this…

That crazy idea was that it should be possible to use some form of pattern recognition to determine whether a particular set of data “looks” like a known pattern of a curve of best fit and an associated error distribution. This would be similar to how face recognition software or spam filtering software works.

I am sure it could be done through a supervised learning method of some sort, where a large number of Monte Carlo simulations are used to derive the training data for any particular combination of curve of best fit and underlying error structure.

But, at the moment I am a newbie to machine learning…  So I am going to need to

Post Revisions:

Written by Jacob Walker

November 13th, 2015 at 11:59 am

Posted in Data Science

Leave a Reply

%d bloggers like this: