Demand forecasting on supply chain using ML and NN

Abstract

Demand forecasting is mainly a process whereby analyzing historical sales data, strategic and operational strategies are devised in order to estimate customer demand. One of the most fundamental aspects of supply chain management is inventory management, its major goal is to cut expenses, boost sales and profits, optimize inventory, and most importantly, promote customer loyalty. The process of extrapolating relevant sales data may be separated into qualitative and quantitative forecasting, with each relying on multiple sources and data sets. When there is previous sales data on certain items and a predetermined demand, the quantitative forecasting approach is employed. It necessitates the application of mathematical formulas as well as data sets such as financial reports, sales, and income numbers, as well as website analytic. The qualitative technique, on the other hand, is based on new technologies, pricing and availability changes, product life cycles, product upgrades and most significantly, the forecasters’ intuition and experience. Machine learning, clustering, time series analysis, neural networks, KNN, support vector regression, support vector machines, regression analysis, and deep learning are some of the approaches used to anticipate demand. A majority of study has gone into improving demand forecasting, which will enhance supply chain sales and profitability. To do that the researchers mainly focused on using machine learning or deep learning as its main methodology and others like support vector algorithm, time series analysis. However, to our best knowledge, only a handful of research is done using hybrid model consists of both deep learning and machine learning as its main methodology. That is why we want to concentrate on using hybrid models to develop dynamically configurable demand forecasting which eventually will give us promising results.