A New(?) Regression Clustering Algorithm

stats.at.home@gmail.com (Justin and Rachel Silverman) — Sun, 13 Aug 2017 00:00:00 +0000

Motivation
My Solution - Hybrid K-Means/Linear-Regression with Transformation
Starting on Boring Simulated Data
Now a more interesting simulated dataset
More realistic, presense of observational noise
Conclusions and future directions

Motivation

I am a fan of the Stack Exchange forums. In particular, I like Cross Validated and Stack Overflow. An interesting question regarding clustering was posted recently. Essentially someone had the following dataset.

plot(Retirees)

Essentially the poster wanted a way of clustering the observations into the “lines” that are fairly easy to observe in the data. I am going to ignore the fact that these lines are actually the result of artifact (e.g., conversion of discrete values to percentages and then plotting the percentages vs. a variable used to calculate the percentages) and just pretend they are real as I think its still an interesting problem. I am actually going to simulate some non-artifactual data and use this as well.

Machine Learning on Statistics @ Home

A New(?) Regression Clustering Algorithm

Motivation