Sales forecasting methods: the use of analytics to forecast demands (CD)

Submitted to Prof. Debjit Roy
Submitted by PGP 2016-2018 batch in 5th term

In everyday business, managers not only have to oversee the day-to-day operations, but also should be able to anticipate the different possibilities in the future so that they can prepare accordingly. From inventory management to coming up with new products and services, it is important that the key decision makers to have the relevant information that can facilitate better decision making process. By using machine learning and forecasting techniques, the store managers can assess the upcoming demands in a much more systematic manner instead of relying solely on intuitions. This report aims to explore and find the advantages of different machine learnings methods to predict the sales. From Simple linear regressions, random forest, ARIMA, to Adaptive Lasso regressions, we aimed to test the forecasting ability of each technique on the data sets of the sales that we acquire from SandwichworkZ. We also identify different factors that are both intrinsic and extrinsic to the sales of the restaurant so that we can build a robust model for forecasting sales not only for the restaurant businesses but also for other businesses in general. From the different machine learning models that we have tested on the data that we have trained, we can see that Random Forest and ARIMA method yields good result in forecasting the sales trends and cycles. It is worth noting that the Adaptive Lasso capture the peaks and lows of the data the best among all the models tested and yields lower mean absolute percentage error (MSPE) than other models. The models studied can help managers understand the different tools available to predict their sales not only to know the upcoming trends but also to make better decisions, reduce preventable losses, and grab the future opportunities.