Given that you have only 5 to 15 minutes, I think you have to decide whether you would rather a) give an overview of the field, which is now overwhelmingly huge or b) give a specific usable case study

I'd go with b), the specific, usable example. I teach R and basic machine learning (and/or how one would use either with trading strategies, etc; depends on the person's interests) over Skype + TeamViewer and what I find is that doing something together on the command line is most engaging. For example, "Here's a CSV of super market sales and clearly, column 4, amount spent by the customer, seems to be predictable based on the time of the day. Let's see a couple of ways we can figure out how!", and dive in. (I use randomForest and talk about variable importance and classification vs. regression. The fact that they can now actually use this ML technique right away on real data, thanks to the robustness of randomForest, is exciting!

These are one-on-one sessions that I am referring to, but I suspect it would scale equally well to a 1:30 classroom situation, too.

Yeah was thinking of giving a quick overview with a not-so-deep dive into the easier stuff like Naive Bayes classification. But your suggestion makes sense as well. I like the supermarket example. Do you maybe have a list of these that I could demo? A link to a specific demo problem?

Thanks a lot

Check out the UCI Machine Learning Repository for a (huge) list of data / problems to test your ML-fu on : http://archive.ics.uci.edu/ml/

library(randomForest) in R comes with some automobile data from the above repository and a small worked example which can be greatly expanded upon. For example, it's fun to try and predict manually whether the car is a 2-door or 4-door by staring at the data, which has 26 columns, so it may be not that obvious:

  > summary(imp85$numOfDoors)
  four  two
   112   81

The variable-importance plot reveals all : the bodyStyle is a key predictor!

  > varImpPlot(numDoors.rf)

I uploaded it here :http://i.imgur.com/IU98gfo.png

  > summary(imp85$bodyStyle)
  convertible     hardtop   hatchback       sedan       wagon
            6           8          63          92          24

Hope this is useful.

Awesome thanks! It is useful

Here is an example of a fun machine learning talk: http://m.youtube.com/watch?v=kKe4M4iSclc

Its a bit longer than 5 minutes but there are some great interactive examples