Forecasting techniques for sales prediction

Abstract

We all are extremely inquisitive about future, eager to comprehend what will occur in the following minute. Correspondingly, retailers are additionally inquisitive about the fate of their business, it’s encouraging and their future deals. Walmart is the world’s biggest retailer and also has a vast grocery chain over the world. It was initially established in America 1962. In 2019, it has more than 11,000 stores in 28 countries but the sales differ from place to place. Many sales strategies, discount rates will be introduced for the improvement of sales. Retailers always try to attract the common people to visit their store. They always focus on improving the future sales. Using some Machine learning forecasting models, we can estimate the future sales based on the past data. Our aim is to apply time series forecasting models to retail sales data, which contains weekly sales of 45 Walmart stores across United States from 2010 to 2012. There are other factors which effects the analysis of weekly sales-markdown, consumer per index, IsHoliday (Boolean value returns whether it is holiday or not), size of the store, unemployment, store type, fuel price and temperature. The forecasting models applied for the data are (Auto-Regressive-Integrated-Moving-Average) ARIMA model and (FeedForward-Neural-Networks) FFNN. The dataset will be divided into training and testing datasets. The predicted values will be checked with the test data and accuracy will be calculated. Based on the accuracy we conclude which of the two models will better for the sales prediction.