Applying Multi-Layer Perceptrons (MLP) for Demand Forecasting in Supply Chain Management

Demand forecasting plays a significant role in supply chain management and business operational effectiveness. This paper focuses on the use of Multi-Layer Perceptrons (MLP) in demand forecasting which is one of artificial neural networks due to its proved ability to model non-linear patterns. This typology suggests a detailed approach which embraces data gathering, data preparation, model building and model assessment. The MLP model is compared with other models including ARIMA, Random Forest and LSTM models. Therefore, the use of MLP model indicated improved performance in terms of RMSE and MAPE in comparison to traditional approaches. Although LSTM has slightly higher performance, it is relatively more simpler and faster to implement MLP which results to better accuracy and fewer computations. It is identified that historical sales data, seasonal factors and economic factors are some of the most important features that affect the accuracy of the forecast. The use of MLP during the implementation of the project proves the efficiency of the proposed approach and the obtained results. This paper reveals the possibility of MLPs in improving the part of demand forecasting in the supply chain management process and suggested ways that could increase opportunities for better supply chain management through the management of inventory. Further research can be dedicated to the elaboration of the combined approaches utilizing the advantages of the notable methodologies for sales forecasting..