2
$\begingroup$

I am currently applying for a Data Science position and have to finish a take-home challenge for one of the companies. However, I don't really understand what they want me to do and hope you can help me interpret the question. Unfortunately, I can't reach anyone at the company due to Christmas holidays

I have a data set with transactions of customers and another dataset with information about age, country... of those customers. Now I am asked to define a target metric for engagement and say how I would define unengaged vs. engaged customer. Afterwards, I have to use this logic to build a model to classify engaged vs unengaged customers (they point out overfitting due to feature used in the target metric).

How would you interpret these questions?

Do I first have to say if target_metric > x: engaged and then build a supervised model to classify the users again?

OR

Would you define several metrics that could separate engaged and unengaged customers and then use an unsupervised model to find two clusters that would then be engaged vs unengaged?

I tried to keep the question as general as possible to avoid "cheating". I only want to know how you would interpret them. Thanks!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

Second one but dont stop there.

First of all you specifying if target_metric > x: engaged is just wrong and you have to let data tell you this x, not you choosing it. Second one seems nice I would start with that, do some extensive analysis and only then start finding x empirically. How? just of the bat you can pose it as minimization problem for different x and find the one that has biggest purity/separation between classes.

Why dont stop there? There is no wright and wrong, if you have some plausible approach in theory but it fails in practice I would definitely include it and elaborate on it, discuss it etc...

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thanks! I will look into the points you raised. I went with the second approach since it makes more sense, just like you said. I found that by transforming two of the engagement metrics I get a very nice separation into two main clusters in a simple kde-plot. I might be on the right track $\endgroup$
    – Blo4d
    Commented Dec 24, 2019 at 14:11

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.