Cognitive Demand Forecasting with Novel Features Using Word2Vec and Session of the Day

Abstract

Demand Forecasting is one of the most crucial aspects in the supply chain business to help the retailers in purchasing supplies at an economical cost with the right quantity of product and placing orders at the right time. The present investigation utilizes a years’ worth of point-of-sale (POS) information to build a sales prediction model, which predicts the changes in the sales for the following fortnight from the sales of previous days. This research describes the existing and newly proposed features for demand forecasting. The motivation behind this research to provide novel features is to obtain an improved and intuitive demand forecasting model. Two features proposed are: Item categorization using word2vec with clustering and session of the day based on the time. The demand forecasting models with traditional features like seasonality of goods, price points, etc. together with our proposed novel features achieve better accuracy, in terms of lower RMSE, compared to demand forecasting models with only traditional features.