.

# Crazy Data Science Tutorial: Classification and Clustering

In order to write a tutorial about classification, it was necessary to find an example that was broad enough that it would need to be sub-divided. Since I actually care about whether you remember this stuff, it needed to be something that a lot of people like and would relate to. And since I have a lot of international subscribers, it needed to be cross-cultural as well. So what is universal, cross-cultural, and dearly loved?

Beer.

There’s American beer, Mexican beer, German beer, Belgian beer….hell, even the Japanese make beer. There’s IPA, Lager, Pilsner. Dark, light, stout. There are so many ways to classify beer that we could spend weeks doing it (so naturally, I did).

Now, before you can classify anything you have to determine the characteristics that you’re going to use. For beer you could use country of origin, color, alcohol content, type of hops, type of yeast, and calorie count among other things. That way you could sort based on any of those characteristics to judge similarities between the various brews.

And just like that, you’ve done classification. Simple, right?

To take the example further, let’s assume that my favorite beer is Sweetwater “Take Two” (a pilsner made here in Atlanta) but I’m in Santiago, Chile this week for a conference.  The Chileans are a lovely people, but the management at my hotel doesn’t know about the wonderful goodness made by Sweetwater and they don’t have it at the lobby bar. I explain my predicament (read: “impending crisis”) to the bartender. What would a good bartender do?

If he’s been in the business for any length of time, he’s already gone through the classification step for beers but probably didn’t realize it. He has them sorted by characteristics in his head. He starts asking me questions about Take Two: “How dark is it?”, “How ‘hoppy’ does it taste?”, and “How many can you drink before passing out?”. Based on my answers he knows that what I’m describing is basically a golden-blonde pilsner with spicy hops and an earthy tone.

He might also figure out that I’ll need help back to my room at the end of the night because of all this “field research”.

So now that he has the characteristics of my favorite brew figured out, he compares that against the beers he knows. The ones with the most matching criteria form a “cluster” that he can make recommendations (and hopefully free samples) from. My night is saved, and his tip is big. Everyone is happy.

And just like that, you understand clustering.

How does this apply to the business world? There are many potential applications of classification and clustering, but a common one is identifying the characteristics of a company’s best customers and then searching a pool of potential customers for ones that meet those characteristics. If your best customers have between 1000-2500 employees, are in the manufacturing and retail verticals, and are located in the New England area of the US, that’s good information to know.

What applications can you think of?

Views: 14006

Comment

Join Data Science Central

Comment by Randal Scott King on August 29, 2014 at 8:11am

True, Wayne. Very True.

Comment by Samson Sani Nzevela on August 29, 2014 at 3:12am

Interesting way of data normalization. Someone needs to carry out some serious classification of Kenyan brand of beer and general alcohol...perhaps an interesting business opportunity to help fix a problem! Google the news on kenya alcohol deaths. Good work.

Comment by Wayne Hart on August 28, 2014 at 2:46pm

There are two clusters in this universe: If it looks like beer, smells like beer and tastes like beer - it is beer.  Everything else is not beer, regards Wayne