Comparison of statistical and machine learning methods for daily SKU demand forecasting
- PDF / 925,268 Bytes
- 25 Pages / 439.37 x 666.142 pts Page_size
- 81 Downloads / 273 Views
Comparison of statistical and machine learning methods for daily SKU demand forecasting Evangelos Spiliotis1 · Spyros Makridakis2 · Artemios‑Anargyros Semenoglou1,2 · Vassilios Assimakopoulos1 Received: 11 June 2020 / Revised: 26 July 2020 / Accepted: 10 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Daily SKU demand forecasting is a challenging task as it usually involves predicting irregular series that are characterized by intermittency and erraticness. This is particularly true when forecasting at low cross-sectional levels, such as at a store or warehouse level, or dealing with slow-moving items. Yet, accurate forecasts are necessary for supporting inventory holding and replenishment decisions. This task is typically addressed by utilizing well-established statistical methods, such as the Croston’s method and its variants. More recently, Machine Learning (ML) methods have been proposed as an alternative to statistical ones, but their superiority remains under question. This paper sheds some light in that direction by comparing the forecasting performance of various ML methods, trained both in a series-by-series and a cross-learning fashion, to that of statistical methods using a large set of real daily SKU demand data. Our results indicate that some ML methods do provide better forecasts, both in terms of accuracy and bias. Cross-learning across multiple SKUs has also proven to be beneficial for some of the ML methods. Keywords Forecasting accuracy · SKU demand · Neural networks · Regression trees · Cross-learning
1 Introduction Daily SKU demand data is typically characterized by irregular demand sizes (erraticness) and variable demand arrivals (intermittency), with many observations having zero values. This is especially true when forecasting at low cross-sectional levels, such as at a store or warehouse level, or dealing with slow-moving items. * Evangelos Spiliotis [email protected] 1
Forecasting and Strategy Unit, School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece
2
Institute for the Future, University of Nicosia, Nicosia, Cyprus
13
Vol.:(0123456789)
E. Spiliotis et al.
Thus, effectively forecasting intermittent, lumpy, erratic, and smooth demand series (Syntetos and Boylan 2005) becomes a challenging task, requiring different methods to those used for extrapolating continuous, regular data (Spithourakis et al. 2011; Petropoulos et al. 2013). However, daily demand forecasting is very common in many industrial and retail settings (Johnston et al. 2003; Willemain et al. 2004; Syntetos and Boylan 2005). Moreover, in many companies, such forecasting must take place for thousands of items and numerous locations (Seaman 2018). Given that inventory management and stock control builds on demand forecasting, with small gains in accuracy leading to considerable inventory reductions (Syntetos et al. 2010) and slight inaccuracies to higher stock holdings and lower service levels than the ones desired (Ghobbar
Data Loading...