Abstract. In this paper, we present a model and methodology for accurately predicting the following quarter’s sales volume of individual products given the previous five years of sales data. Forecasting product demand for a single supplier is complicated by seasonal demand variation, business cycle impacts, and customer churn. We developed a novel prediction using machine learning methodology, based upon a Dense neural network (DNN) model that implicitly considers cyclical demand variation and explicitly considers customer churn while minimizing the least absolute error between predicted demand and actual sales. Using parts sales data for a supplier to the oil and gas industry across North America, we found our novel method to predict demand with the least mean absolute error(MAE) of 0.65 and a coefficient of determination(R 2 -score) of 0.90. Dense neural network model is our solution to forecast demand after being compared with Timeseries ARIMA, and regression models
Kosinovsky, Hannah; Daggubati, Sita; Ramasundaram, Kumar; and Allen, Brent
"A Data Driven Approach to Forecast Demand,"
SMU Data Science Review: Vol. 2:
3, Article 1.
Available at: https://scholar.smu.edu/datasciencereview/vol2/iss3/1
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License