Forecasts Walmart sales based on weather data.
Using data from a Kaggle competition, I predicted sales of given items at Walmart locations based on weather. Abstract is below; for more details, see ProjectReport.pdf. Project files here include python scripts but not dataset. I created this as my final project for CS 491/591 at the University of New Mexico in Fall 2015.
Abstract—Retailers must forecast customer demand in order to stock the right products in the right quantities. Companies like Walmart analyze sales data to prevent “a retailer's twin nightmares: too much inventory, or not enough” [1]. For a previous Kaggle competition, Walmart released a dataset for 45 stores over 15 months [2]. The data included the daily units sold at each store of 111 items for which demand might vary with the weather, such as milk or umbrellas. Data from the weather station nearest each store was provided as well. Using a collaborative filtering approach, I sought to predict sales of each item at each store based on the daily average temperature. I found the most accurate results by assigning each temperature to a range, then using the range as the basis of the predictions.
[1] C.L. Hays (2004, Nov. 14). “What Wal-Mart Knows About Customers' Habits”. New York Times.
[2] Kaggle. “Walmart Recruiting II: Sales in Stormy Weather,” kaggle.com.