![]() ![]() We explore extensions of these models in several directions, which are natural consequences of our new framework to systematically tackle IDF. We start working from Croston’s original insight, towards a consistent set of flexible intermittent demand models. In this paper, we make a sequence of observations on IDF methods proposed so far. Towards developing forecast distributions and uncertainty estimates, statistical models underlying Croston’s method were proposed by. Upon recognizing that traditional approaches such as simple exponential smoothing (SES) led to poor forecasts in slow-moving inventory, Croston proposed to independently apply exponential smoothing to consecutive positive demand sizes, and to the number of periods between each ( i.e., interarrival or interdemand times). IDF was recognized as a unique and challenging problem in the early 70s. Therefore, precise estimates of forecast uncertainty, e.g., with access to forecast distributions, are vital for IDF. For example, spare parts in aerospace and defense are well known to exhibit intermittent patterns. Intermittent demand is most likely to appear with slow-moving, (sometimes) high-value items that are critical to production processes. ĭemand for large shares of inventory catalogues in manufacturing are well known to exhibit intermittency. Not only does this sparsity render most standard forecasting techniques impractical it leads to challenges on measuring forecast accuracy, model selection, and forecast aggregation. Intermittent demand forecasting (IDF) is concerned with demand data where demand appears sporadically in time, i.e., long runs of zero demand are observed before periods with nonzero demand. This does not alter our adherence to PLOS ONE policies on sharing data and materials. The specific roles of these authors are articulated in the ‘author contributions’ section.Ĭompeting interests: The first three authors of this article (ACT, TJ, and YW) are employed by Amazon Web Services. The funder provided support in the form of salaries for authors ACT, TJ, YW, but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. įunding: The first three authors of this article (ACT, TJ, and YW) are employed by Amazon Web Services. RAF and Auto data sets are available online at. The M5 data set is available on Kaggle ( ). The Car Parts data set is available in the Monash Time Series Forecasting Repository ( ). This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.ĭata Availability: The UCI data set is available on the UCI ML repository ( ). Received: SeptemAccepted: OctoPublished: November 29, 2021Ĭopyright: © 2021 Türkmen et al. Complementing this theoretical advancement, we demonstrate the efficacy of our framework for forecasting practice via an extensive empirical study on standard intermittent demand data sets, in which we report predictive accuracy in a variety of scenarios.Ĭitation: Türkmen AC, Januschowski T, Wang Y, Cemgil AT (2021) Forecasting intermittent and sparse time series: A unified probabilistic framework via deep renewal processes. This leads to more flexible modeling in scenarios where data of individual purchase orders are directly available with granular timestamps. We also demonstrate that modeling continuous-time demand arrivals, i.e., with a temporal point process, is possible via a trivial extension of our framework. The connection to discrete-time renewal processes allows not only for a principled extension of Croston-type models, but additionally for a natural inclusion of neural network based models-by replacing exponential smoothing with a recurrent neural network. Our framework is based on extensions of well-established model-based methods to discrete-time renewal processes, which can parsimoniously account for patterns such as aging, clustering and quasi-periodicity in demand arrivals. We introduce a new, unified framework for building probabilistic forecasting models for intermittent demand time series, which incorporates and allows to generalize existing methods in several directions. Intermittency are a common and challenging problem in demand forecasting. ![]()
0 Comments
Leave a Reply. |