Short term forecasting for lumpy and non-lumpy intermittent demands

Abstract

Accurately forecasting intermittent demands is a concern to many industries. This paper proposes an approach to improve forecast accuracies on intermittent demands given up to 36 months of historical data. The conventional approach to forecasting problems with irregular patterns is Crostonpsilas method. We use different methods based on modifications of Crostonpsilas method to forecast lumpy intermittent demand and non-lumpy intermittent demand. The historical data for lumpy intermittent demand is split into three series while that for non-lumpy demand is split into two. Forecasting is then performed separately on each of the series. The intermittent demand forecaster has been tested on two datasets and compared to Crostonpsilas method. The intermittent demand forecaster is able to reduce average forecasting error by 10.22% and 27.42% compared to Crostonpsilas method for non-lumpy demand and lumpy demand, respectively.