One class label learning

Hi everyone,

Just starting out using Rapid Miner and I've hit a snag which I can't seem to pass.

Basically, I have a data set which contains all the members who churned in the past year.

I then have a seperate dataset which contains all current members and we want to look at whether any of these have the characteristics of the churners from previous years and predict whether they will renew or churn. However, I am currently running into a problem as I only have the one label in the first data sheet which is churned, which then throws up a one class label error.

Any help would be greatly appreciated.

Many thanks,

Rob

Find more posts tagged with

AI Studio

Getting Started

Accepted answers

Telcontar120

It sounds like you should be able to combine your two datasets, using current customers who have been open for some minimum period of time, and then assign them the label of "loyal," as well as the prior churners. That will then give you a dataset with two label values. This should then be suitable for a wide variety of different learning algorithms and is a classic predictive modeling project.

All comments

Telcontar120

rob

Thanks!

Thomas_Ott

@rob not to toot my own horn but I have a sample process that does something similar here: http://www.neuralmarkettrends.com/blog/entry/use-rapidminer-to-auto-label-twitter-training-set

JEdward

Warm memories, my very first production RapidMiner model was a one-class SVM. The use case was that we knew all the purchasers of a particular product, but we didn't know who had or had not been previously offered it so we couldn't build a "buy/not buy" model.

Instead we used one-class SVM to model those previous purchasers and then use that model to decide who and how to send marketing about the product. In the 1st year it was in production it generated a couple of million pounds of additional revenue - so I definitely recommend it as an approach.

Thomas_Ott

@JEdward that sounds spectacular! The one class SVM has some great uses if you're trying to better than a "shot in the dark" so to speak!