Defining Artificial Neural Network Hyperparameters for Forecasting Nonstationary Demand of Spare Parts

The supply chain management (SCM) approach's main purpose is to maximize profitable growth. Due to its influence on supply chain decision-making, forecasting is essential to attaining this goal. The nonstationary demand patterns are the most prevalent and difficult behavior seen in the supply chain, particularly for spare parts. Numerous forecasting methodologies have been created in various sectors due to the significance of the predictions and the hard demand behaviors linked with them. Statistical forecasting methods, which rely on past trends to project the future, are one category of these forecasting approaches. Croston's methods and autoregressive integrated moving average (ARIMA) are the two statistical forecasting techniques that are most frequently employed for non-stationary demand. However, it has been shown that these methods have several weaknesses, especially when dealing with highly uncertain data. In this case, smoothing the data is required prior to applying these forecasting strategies. Recently, Artificial Intelligence (AI) approaches such as Artificial Neural Network (ANN) are emerging in the field of time series forecasting that is used to build estimation algorithms. ANN-based forecasting algorithms have demonstrated the ability to outperform traditional statistical approaches. However, ANN models can only function effectively if their hyperparameters are properly chosen.