2
$\begingroup$

Customer Segmentation and Category association I have to solve two questions on the following dataset: 1. arrange customers into mutually exclusive groups.explain the clusters. 2.identify 1-1 product category association rules for each cluster, i.e. if a customer bought from this category, they are likely to buy from this category too.

$\endgroup$
2
  • 1
    $\begingroup$ you are actually looking for market basket analysis for number 2. look up things like conviction and all the related metrics that tell you which items are commonly purchased with each other. $\endgroup$
    – Victor Ng
    Mar 2, 2020 at 15:34
  • $\begingroup$ thank you for the suggestion! i will try to implement that $\endgroup$ Mar 3, 2020 at 1:09

2 Answers 2

0
$\begingroup$

Question 1.:

Encode the columns, (Labelencoding, tf-idf, feature generation etc...)<-> quantify the columns, than iterate fastly with different algos and evalute and compare the results, for example with silhuette score.

Question 2. Once you identified the best clustering algorithm, look for the categories that (uniquely or very likely) determine certain cluster/group. It wont be 1-1 but in probabilities determined.

$\endgroup$
1
  • $\begingroup$ thank you for the suggestion! I was trying to use the clustering algorithm but i find it a bit difficult as it doesn't have any product ratings. So i have a cold start $\endgroup$ Mar 3, 2020 at 1:10
0
$\begingroup$

Question 1: You have 2 options. Since you are working with Python, all your categorical features will need to be dummy coded anyway. If you consider your zeroes and ones as numeric entries after the dummy coding, you may go for an algorithm like k-means clustering that only takes in numeric data. For mixed datasets, partition around medoids is also a useful algorithm. You may use Manhattan distance instead of Euclidean distance as in case of k-means to highlight the differences across observations. Alternatively, you may try out hierarchical clustering which should also provide good results. The evaluation of results can be done by looking at the silhouette plot and silhouette index.

Question 2: Apriori algorithm helps you come up with association rules that can help out in this case as well.

$\endgroup$

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.