Hi !!!

 I get a list of 10000 customers nad asked to predict who will by new product of company.

How I can attack this business problem ? IF I need to use clustering ?

You can only predict something if you have data about how it has behaved in the past.  So you need to think about "what kind of customer purchases this new product" and find something similar, that you do have data for. 

 Is the new product similar to an existing product that people have bought?  Then model propensity to buy the existing product.

Are people who usually buy from the company likely to also buy this new product?  Then model propensity to buy any product, or expected time to next product purchase.

Do you know the characteristics of people likely to purchase the new product?   Then find them in the data.

If you don't know any of this, use clustering to break your customers into groups.  Then profile the clusters and pick the cluster that you think is most likely to buy the product.

If you want to do this properly, offer the new product to a random sample of customers, and use their purchase decisions to build a buyer profile and a propensity model of who will buy the new product.  Clustering is not suited to your business problem.  It would be better to run a test campaign and obtain appropriate data.

Thank you for your answer.

 What do you mean "build a buyer profile and a propensity model " ? Is it decision tree ?

Tree with leafs defined as "buyer" (in opposite to "non-buyer)

describes the profile and good tree

can be used for prediction ?

May by I refrase . How from profile I can to predict ?

Thank you

 

Yes, do a decision tree.  You can use it for prediction, and it will be the first part of a buyer profile.

The buyer profile helps people understand what kind of customers buy the new product, so that they can make better offers and advertisements.  I think this will be helpful in your project.

THank you . I do it . I showed partial results to my customer . There is some attribute that he say must be in model (decision tree) in rules . But the attribute is on level 6-7 in tree when I build very big tree. How can I to convince him that this attribute isn't so important . I consider calculating corelation between this attribute and target atribute , building tree with this attribute only and show him that model accuracy reducing drastically , in addition I acn draw for him simple graph target attribute vs. this attribute to show him that for each value of this attribute the target value (2 at all) distribute equally (50-50) .

 Have you more ideas regarding the problem , more right , may be . I have considered the concept of attribute selection but I think it will take much time to explan this term to my customer .

Do the chart.  If he doesn't understand attribute selection then he won't understand correlation either.  Show him pictures instead.