A retail firm “ACME Stores Corporation” would like to profile behavioral customer purchase information to understand the purchase basket size in USD (spend_amount) of several items across the product taxonomy of the company. The managers have collected a sample of various customers for high rotation articles over the last month. The dataset has been enriched with additional information coming from the company CRM. This data pertains to sociodemographics (age, gender, marital status, city class and place of residence duration), product details (product SKU and category) and spend_amount from last month. At this point in the project, they would want you to calibrate a model to forecast the spend amount of each customer across several products. They expect this special forces project to help them launch a marketing campaign of hypersegmented offerings for the customers of the company.
Your model performance will be evaluated based on the predictive ability over the spend_amount variable in an external data set, which has been withheld from you by your evil manager (see picture below). Your job is to create a prediction archive according to the rules that will be given by the instructor in class. Forecasts will be measured against the root mean squared error metric (RMSE). Bear in mind that RMSE inflicts larger penalties on large deviations, so it is in your best interests to not fail spectacularly.
Business Glossary
Variable
Description
Customer_ID
Unique Customer identifier (anonymized)
Product_SKU
Product stock keeping unit (anonymized)
Gender
Sex of Customer under consideration.
Age
Age of customer, discretized in ranges.
Job
Customer job (masked)
City_Type
Type of the City (according to size)
Residence_Duration
Number of years since they moved to their home location
Marital_Status
Whether the customer is married or not
Main_Category
Product class within the company taxonomy (anonymized)
Secondary_Category
Additional category the product may belong (anonymized)
Tertiary_Category
Additional category the product may belong (anonymized)
Spend_Amount
USD spent for this product by this customer (Outcome attribute to be forecasted)