Improving intermittent demand forecasting based on data structure

Forecasting spare parts requirements is a challenging problem, because the normally intermittent demand has a complex nature in patterns and associated uncertainties, and classical forecasting approaches are incapable of modeling these complexities. The present study introduces a hybrid model that can impressively overcome the limitations of classical models while simultaneously using their unique advantages in dealing with the complexities in intermittent demand. The strategy of the proposed hybrid model is to use the three individual autoregressive moving average (ARMA), single exponential smoothing (SES), and multilayer perceptron (MLP) models simultaneously. Each of them has the potential of modeling a different structure and patterns of behavior among the data. The accuracy in forecasting ability is also increased by the suitable examination of these in the intermittent data. Croston’s method is the backbone of the suggested model. The proposed hybrid model is based on CV² and ADI criteria, which improve its efficacy in examining inappropriate structures by reducing the cost of inappropriate modeling while increasing the prediction model accuracy. Using these results prevents the hybrid model from being confused or weakened in the modeling of all groups and reduces the risk of choosing the disproportionate model. The accuracy of prediction models was evaluated and compared using mean absolute percentage error (MAPE) by implementing an example, and promising results were achieved.