Blazers: Breaking It Down (Part II)

(Check out Part I of my Blazer Posts here)

One of the luxuries of having really small data sets is that it makes it easy to play around with different ways of analyzing them. This was particularly true here, because it took me a while to figure out the best way to analyze my blazers. More than with some of my earlier analysis, I really wanted to know what I should look for in future blazer purchases vs. just what I should get rid of. First, I tried grouping them by color, which wasn't super informative. (hint: I own a lot of black) Then, I tried building a matrix looking at them by color and style, which was a little more helpful but still only gave me two dimensions to look at the problem.

In my data mining class, we've been talking about using classification trees in analyzing data. On a very basic level, classification trees help you "classify" different items based on their "features" or "attributes". Here, my items are my blazers and I picked three "attributes" that I could use to classify them: Color (Patterned or Solid), Style (Collar or Collarless), and Suit (Part of a Suit or Not)

This turned out to be surprisingly helpful and I ended up with a pretty picture. I knew that I had a lot of solid colors (even the patterned one kind of looks solid from a distance) and that a lot of my blazers seemed to pretty similar in style. But by constructing this tree, I was able to see which were the most important features for each blazer. For example, given how many solid blazers I have, it's more important to know my French Connection blazer is patterned than it is to know that it's got a collar or is part of a suit. Because it's different (and happens to be highly rated), it's both an item I should keep and might be an example of an area where I could add more items to my closet (patterned blazers galore!)

What does this actually mean for my shopping habits? Well, there are a couple of ways to interpret this data. Someone who had no experience with women's closets or any personal knowledge of me, might use the last branch of the tree to suggest that I go for a suit blazer but not buy the rest of the suit . . . or only shop for used suit blazers at thrift stores. That might be a little tongue-in-cheek, but it does show that you definitely need context when interpreting data!  In fact, those happen to be my oldest blazers, and at least two of them are on their last leg. More likely, one of two things are going on, either:

a) I've figured out what styles/colors works for me and I 
should just stick with them, buying more of the same

OR

b) My closet could use a little diversity 
(I'm leaning towards "b"). 

If I went with option "b", the features I have on my tree present different ways I could add a little more pizzazz to my wardrobe. I could choose to diversify based on cut. Or, if the suit style is really best for me, I could choose to add more patterned pieces. Or, I could be really wild and go for both. 

Zara Printed Kimono Blazer

Boom. Sold. (except that pattern gives me a headache)

* * * 

*Please note, that classification trees are normally built out using much more rigorous techniques. I used the idea more as inspiration rather than actually building a mathematical model. However, you could say, in theory, I used some sort of majority classifier approach. Another thing to know is that the patterned blazer is both part of suit, and a collared style. But because I started with color as the first branch of my tree and there was only one patterned blazer, color ended up being the most important attribute for this particular instance.