Retail Incident Prediction

Welcome to the Retail Incident Prediction Marathon Match!

Our client for this match is a retail chain interested in leveraging data science techniques and machine learning to forecast maintenance incidents in their stores based on available historic data. We will focus on the six most frequent incident types, labeled as cash_drawer, pin_pad, register, coffee, frozen_drink, and soda_fountain As the label names suggest, the first three indicate different malfunctions of the store counter equipment, and the latter three refer to malfunctions of coffee, frozen-drinks, and soda vending machines installed in the stores. The following data are available as inputs (training datasets):

• The log of past incidents, including store location ID, incident type, timestamp, duration, and other details.

• The daily log of relevant transactions in past, with a per-store breakdown. It includes coffee, soda, and frozen drink purchases in vending machines, and overall sales in each store. Beware, available data are sparse, i.e. for some stores and days the data may be missing, though for each store and transaction type it is guaranteed that data are present for at least 15 days in every month covered by the dataset.

• The aggregated daily transaction data for entire training and forecast period, without per-store breakdown.
  
  As you will see, the dataset we are working with covers at most eight months of data (in the final test, 6 months of training data + 2 months to forecast). At the same time, the retail sales, especially those of consumables like soda and cold drinks, are clearly season-dependent. The best workaround we have is to provide the aggregated transaction information for entire training + forecast period. The aggregation was done by summing up transactions over different stores. You may think of it as data about generic transaction trends over a year.

Based on these data you are asked to forecast incident numbers and average durations in each store in the period following provided log data.

Match rules to be determined