Alberto Puentes

The goal of our capstone project was to leverage historical Walmart sales data and macro economic indicators into a predictive forecasting model that would enable a stakeholder to accurately forecast sales demand one week in advance. With supply chains currently stressed, the timing of our project findings could help logistic managers accurately forecast demand and make appropriate inventory decisions. In our project we converted Walmart time series data into a regression problem and identified key weekly sales drivers through various feature manipulations and engineering. We constructed multiple machine learning models but ultimately found that our Polynomial Regression model which leveraged a Recursive Feature Elimination model performed the best. Based on our RMSE scores, we were able to outperform our Baseline (predictions equal to last year’s sales numbers) by roughly 29% with an R-squared of 0.987.