Code for the Kaggle repeat-customer acquisition challenge

Features

For each training instance (an offer given to a customer) we will calculate the following features:

• Data already available
  • chain_id - the store the customer frequents
  • market_id - the region that the store is located in
• Transaction analysis
  • Customer's behaviour
    • num_chains_visited_lastX Number of unique chains visited in the past x days
    • num_visits_lastX Number of total visits in the past x days
    • avg_spend_per_visit_lastX Average spend per visit in the past x days
    • amt_purchases_lastX Total spend in the past x days
    • avg_price_lastX Average price of item purchased in the past x days
    • num_categories_shopped_lastX Total number of categories shopped in the past x days
  • Customer's past activity in this category
    • amt_category_purchases_lastX Total customer spend in this category in the past x days
    • pct_category_of_total_wallet_lastX Category share of wallet for this customer in the past x days
    • amt_category_purchases_lastX Quantity purchased in this category in the past x days
    • num_category_visits_lastX Number of visits (unique chain-date) in this category in the past X days
    • `pct_category_visits_of_total_lastX) Number of visits in this category as % of total in the last X days
  • Customer's past history with brands in this category
    • amt_category_brand_purchases_lastX Total customer spend on this brand in this category in the past X days
    • pct_category_brand_total_wallet_lastX Share of customer wallet in this category that this brand holds
    • num_unique_brand_purchases_in_category_lastX Number of other brands customer has purchased from in this category
    • (Could do similar for company instead of brand)
  • Customer's past history with this brand in other categories
  • Customer's history with this department
  • Propensity to develop loyalty
    • Avg unique products purchased per category
  • Offer timing
    • Month offered
    • % of annual demand for brand in month offered
    • % of annual demand for category in month offered

Things to consider

• offer_date - chunk it based on time (e.g. month, week or day of week?)
• Properties of the underlying demand for the offered brand - e.g. if it's a seasaonal product we may get higher conversion rates when in-season