2
$\begingroup$

I was just thinking how ML techniques can be applied in the retail industry. Suppose we have data from a retailer who deals with apparel and cloth in this format and for each item there are pre-defined features, for example a shirt will have features like color, half/full sleeves etc.. I want to understand how to extract important features of "most sold items" from this data. For example if I find that shirts are the most commonly sold item, then which kind of shirt : Black in color, Half Sleeves, etc...

I was thinking to use decision trees here. I am not sure if this is a good approach.

Improve this question
$\endgroup$
2
  • $\begingroup$ Hi Amit; I made a couple of edits to clarify your post, but I was not sure what to do with "apparel and cloths", which in this context might be "apparel and cloth" (cloth being a mass noun, doesn't gain a plural in this context, in the same way that 'rice' doesn't), or it might perhaps have been intended as "apparel and clothing". I assumed the first, but please alter it if I misunderstood. $\endgroup$
    – Glen_b
    Commented May 29, 2013 at 4:17
  • $\begingroup$ Why do you need such a model? I ask this because many shops now lack diversity, probably since they use some statistical methods to determine the most commonly sold items and then sell only these items. As a result, the market immediately becomes oversaturated with these items. $\endgroup$
    – user31264
    Commented Sep 28, 2014 at 18:30

2 Answers 2

Reset to default
2
$\begingroup$

Lasso seems like it would be good. Lasso (Least Absolute Shrinkage and Selection Operator) is like OLS, but it penalizes the "length" of the final coefficients by including a "regularizer" term.

The actual paper is Tibshirani (1996).

The Lasso package and demo in R is http://cran.r-project.org/web/packages/genlasso/vignettes/article.

The lasso is great because it penalizes a large number of coefficients to be exactly zero, and you can adjust the tuning parameter to see what drops out first and stays as you become more strict about learning relevant features.

EDIT:

But as the commenter said, be careful of what you are actually learn from a model. Additionally, if you are a larger firm and begin to change your production...be aware of https://en.wikipedia.org/wiki/Campbell%27s_law

Improve this answer
$\endgroup$
0
$\begingroup$

Just get the data, decide how you want to define feature groups (colours per shirt type, or shirt type seperately, colors seperately, etc' ) and encode thema s features using sci-kit learn.

http://scikit-learn.org/stable/modules/preprocessing.html#encoding-categorical-features

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.