Single-hidden layer neural networks for forecasting intermittent demand

摘要: Abstract Managing intermittent demand is a vital task in several industrial contexts, and good forecasting ability fundamental prerequisite for an efficient inventory control system stochastic environments. In recent years, research has been conducted on single-hidden layer feedforward neural networks, with promising results. particular, back-propagation adopted as gradient descent-based algorithm training networks. However, when managing large number of items, it not feasible to optimize networks at item level, due the effort required tuning parameters during stage. A simpler faster learning algorithm, called extreme machine, therefore proposed literature address this issue, but never tried demand. On one hand, extensive comparison trained by improve our understanding them predictors other also worth testing machines context, because their lower computational complexity generalisation ability. paper, are compared benchmark well standard methods real-time series, combining different input patterns architectures. statistical analysis then validate best performance through aggregation levels. Finally, some insights practitioners presented potential implementation real