Demand forecasting with machine learning algorithm: Implementation in a production facility

With the advancement of technology, new concepts emerge and the amount and variety of data is increasing day by day. It has become important to manage this increasing amount and diversity of data. With the increasing competition in the production sector, it has become critical to respond quickly to demands as well as to manage elements such as quality, cost and efficiency. It seems possible to respond quickly to demands with effective planning. Prediction is the basis of an effective planning. It is possible to make predictions with the historical data we have. However, classical prediction methods are insufficient to make these predictions with the increasing variety and amount of data. Within the scope of this thesis, weekly material input and daily material input prediction were made for three different locations of a production facility by using the Support Vector Machines algorithm in machine learning techniques built on strong statistical methods. The weekly forecast yielded better results than the daily forecasts, with MAPE error rate being 25.03% for Çorlu, 6.69% for İzmit and 11.58% for Gebze. Keywords: Demand Forecasting, Machine Learning, Support Vector Machine, Industry 4.0