Need help in transforming a customer structure to an analytical structure

Question

I have the following tables. I am interested in studying the manager's attributes that will predict the sales of Product A. In addition, I want to study whether the client attributes have an impact of the sales of product A. Can someone give me an example on how to combine these three tables so that I can do my analysis?
Sales Managers -> (1-n) Clients -> (1-n) Products
Sales Managers Attributes
Manager ID
Region
State
Years of experience
Age
Clients Attributes
Client Type (Technology / Finance, etc)
Client Size (Large / Medium / Small)
Product Attributes
Products (A / B / C)
Revenue
Sale Date
1. If I transform the data into the following structure, will this structure allow me to study the relationship between the attributes of sales managers/clients and the likelihood of selling product A?
My concern is that each sales manager is repeated multiple times. For example if a sales manager have clients of Technology and of Finance and clients of all sizes (Large/Medium/Small). That's six rows of data for each unique manager.
Manager ID, Region, State, Years of Experience, Age, Client Type, Client Size, # Clients, Revenue, Sale Year, # Products Sold, Product A Sold (yes/no)
123, East, MA, 5, 25, Technology, Large, 2, $10000, 2007, 1000, Yes
123, East, MA, 5, 25, Technology, Medium, 5, $6000, 2007, 500, Yes
123, East, MA, 5, 25, Technology, Small, 10, $4000, 2007, 200, No
123, East, MA, 5, 25, Finance, Large, 1, $5000, 2007, 500, Yes
123, East, MA, 5, 25, Finance, Medium, 1, $5000, 2007, 500, Yes
123, East, MA, 5, 25, Finance, Small, 1, $5000, 2007, 500, No
124, West, CA, 10, 35, Technology, Large, 2, $10000, 2007, 1000, No
124, West, CA, 10, 35, Technology, Medium, 5, $6000, 2007, 500, Yes
2. What algorithm should I use to study the input variables to predict the likelihood of selling product A?

Answer 1

HI,
You could use Association Rules for coming up with the probability of sales given the sale of a specific item.
ODM will handle transactional input of sales information for this purpose.
As for data preparation, you should take a look at Data Miner Workflow's Aggregation Node. This node allows you generate "nested tables' of the type DM_NESTED_NUMERICAL and DM_NESTED_CATEGORICAL. Using this approach, you can create structures to capture the 1 to many (as in your case of sales managers to clients). However you can not use it to capture 1 to many to many ( as in your sales managers to clients to products). See simple script below showing how nested data is creates using the SH.SALES sample data.
For demos on how to do either AR or Aggregration, you can download SQL Dev 3.0 and use it against a DB of 11.2.0.1 or higher. There are OBE (Oracle By Example, available on OTN) tutorials and Cue Card Tutorials (shipped in SQL Dev).
Thanks, Mark
WITH
/* START OF SQL FOR NODE: SALES */
"N$10071" as (select /*+ inline */ "SALES"."PROMO_ID",
"SALES"."TIME_ID",
"SALES"."CHANNEL_ID",
"SALES"."CUST_ID",
"SALES"."PROD_ID",
"SALES"."QUANTITY_SOLD",
"SALES"."AMOUNT_SOLD"
from "SH"."SALES" )
/* END OF SQL FOR NODE: SALES */
/* START OF SQL FOR NODE: Aggregate */
"N$10072" as (SELECT t0."CUST_ID",
t0."QUANTITY_SOLD_COUNT",
t1."AMOUNT_SOLD_COUNT",
t2."QUANTITY_SOLD_SUM",
t3."AMOUNT_SOLD_SUM",
t4."AMOUNT_SOLD_SUM_1"
FROM
(SELECT "CUST_ID", COUNT("QUANTITY_SOLD") "QUANTITY_SOLD_COUNT"
     FROM "N$10071"
     GROUP BY "CUST_ID") t0,
(SELECT "CUST_ID", COUNT("AMOUNT_SOLD") "AMOUNT_SOLD_COUNT"
     FROM "N$10071"
     GROUP BY "CUST_ID") t1,
(SELECT "CUST_ID", SUM("QUANTITY_SOLD") "QUANTITY_SOLD_SUM"
     FROM "N$10071"
     GROUP BY "CUST_ID") t2,
(SELECT "CUST_ID", SUM("AMOUNT_SOLD") "AMOUNT_SOLD_SUM"
     FROM "N$10071"
     GROUP BY "CUST_ID") t3,
(SELECT "CUST_ID",
     CAST(COLLECT(DM_Nested_Numerical(ID5, "AMOUNT_SOLD_SUM_1")) AS DM_Nested_Numericals ) "AMOUNT_SOLD_SUM_1"
     FROM
          (SELECT "CUST_ID",
          ("PROD_ID") ID5,
          SUM("AMOUNT_SOLD") "AMOUNT_SOLD_SUM_1"
          FROM "N$10071"
          GROUP BY "CUST_ID", (("PROD_ID")) )
     GROUP BY "CUST_ID") t4
WHERE t0."CUST_ID" = t1."CUST_ID"(+)
AND t0."CUST_ID" = t2."CUST_ID"(+)
AND t0."CUST_ID" = t3."CUST_ID"(+)
AND t0."CUST_ID" = t4."CUST_ID"(+)
/* END OF SQL FOR NODE: Aggregate */
select * from "N$10072"