Rights statement: The final, definitive version of this article will be published in the Journal, International Journal of Production Economics 143 (1), 2013, © ELSEVIER.
Submitted manuscript, 217 KB, PDF document
Final published version
Research output: Contribution to Journal/Magazine › Journal article › peer-review
Research output: Contribution to Journal/Magazine › Journal article › peer-review
}
TY - JOUR
T1 - Intermittent demand forecasts with neural networks
AU - Kourentzes, Nikolaos
N1 - The final, definitive version of this article will be published in the Journal, International Journal of Production Economics 143 (1), 2013, © ELSEVIER.
PY - 2013/5
Y1 - 2013/5
N2 - Intermittent demand appears when demand events occur only sporadically.Typically such time series have few observations making intermittent demand forecasting challenging. Forecast errors can be costly in terms of unmet demand or obsolescent stock. Intermittent demand forecasting has beenaddressed using established forecasting methods, including simple moving averages, exponential smoothing and Croston’s method with its variants. Thisstudy proposes a neural network (NN) methodology to forecast intermittenttime series. These NNs are used to provide dynamic demand rate forecasts,which do not assume constant demand rate in the future and can captureinteractions between the non-zero demand and the inter-arrival rate of demand events, overcoming the limitations of Croston’s method. In order tomitigate the issue of limited fitting sample, which is common in intermittentdemand, the proposed models use regularised training and median ensemblesover multiple training initialisations to produce robust forecasts. The NNsare evaluated against established benchmarks using both forecasting accuracy and inventory metrics. The findings of forecasting and inventory metricsare conflicting. While NNs achieved poor forecasting accuracy and bias, allNN variants achieved higher service levels than the best performing Croston’s method variant, without requiring analogous increases in stock holdingvolume. Therefore, NNs are found to be effective for intermittent demandapplications. This study provides further arguments and evidence againstthe use of conventional forecasting accuracy metrics to evaluate forecastingmethods for intermittent demand, concluding that attention to inventorymetrics is desirable.
AB - Intermittent demand appears when demand events occur only sporadically.Typically such time series have few observations making intermittent demand forecasting challenging. Forecast errors can be costly in terms of unmet demand or obsolescent stock. Intermittent demand forecasting has beenaddressed using established forecasting methods, including simple moving averages, exponential smoothing and Croston’s method with its variants. Thisstudy proposes a neural network (NN) methodology to forecast intermittenttime series. These NNs are used to provide dynamic demand rate forecasts,which do not assume constant demand rate in the future and can captureinteractions between the non-zero demand and the inter-arrival rate of demand events, overcoming the limitations of Croston’s method. In order tomitigate the issue of limited fitting sample, which is common in intermittentdemand, the proposed models use regularised training and median ensemblesover multiple training initialisations to produce robust forecasts. The NNsare evaluated against established benchmarks using both forecasting accuracy and inventory metrics. The findings of forecasting and inventory metricsare conflicting. While NNs achieved poor forecasting accuracy and bias, allNN variants achieved higher service levels than the best performing Croston’s method variant, without requiring analogous increases in stock holdingvolume. Therefore, NNs are found to be effective for intermittent demandapplications. This study provides further arguments and evidence againstthe use of conventional forecasting accuracy metrics to evaluate forecastingmethods for intermittent demand, concluding that attention to inventorymetrics is desirable.
KW - Intermittent demand
KW - Neural Networks
KW - Croston's Method
KW - Forecasting
KW - Slow moving items
U2 - 10.1016/j.ijpe.2013.01.009
DO - 10.1016/j.ijpe.2013.01.009
M3 - Journal article
VL - 143
SP - 198
EP - 206
JO - International Journal of Production Economics
JF - International Journal of Production Economics
SN - 0925-5273
IS - 1
ER -