Demand Forecasting Intermittent and Lumpy Time Series: Comparing Statistical, Machine Learning and Deep Learning Methods

Abstract

Forecasting intermittent and lumpy demand is challenging. Demand occurs only sporadically and, when it does, it can vary considerably. Forecast errors are costly, resulting in obsolescent stock or unmet demand. Methods from statistics, machine learning and deep learning have been used to predict such demand patterns. Traditional accuracy metrics are often employed to evaluate the forecasts, however these come with major drawbacks such as not taking horizontal and vertical shifts over the forecasting horizon into account, or indeed stock-keeping or opportunity costs. This results in a disadvantageous selection of methods in the context of intermittent and lumpy demand forecasts. In our study, we compare methods from statistics, machine learning and deep learning by applying a novel metric called Stock-keeping-oriented Prediction Error Costs (SPEC), which overcomes the drawbacks associated with traditional metrics. Taking the SPEC metric into account, the Croston algorithm achieves the best result, just ahead of a Long Short-Term Memory Neural Network.

Maritime Spare Parts Management: Current State-of-the-Art

“…As a consequence, forecasting methods should be adapted to such characteristics. In fact, we find some adequate methods within statistical as well as machine learning as well as heuristics dealing with such demand patterns [6,11,15,24].…”

An Empirical Comparison of the Sales Forecasting Performance for Plastic Tray Manufacturing Using Missing Data

“…Nowadays, the time series analysis is widely applied in various sectors, including consumption and businesses [5,9,10,22,28,29], demand forecasting and supply chains [1][2][3]6,7,11,12], economics [30], industrial applications [4], traffic and automatic system controls [8,21,31], meteorology and the environment [17], epidemiology [19,20,[23][24][25], and others [26,32,33]. In the above studies, the authors have proposed assumptions regarding time series data, the corresponding modeling methods, and evaluation indicators for various contexts.…”

“…The classic time series forecasters are based on statistical learning and include the Naive forecasting [4,7,50], the moving average [2][3][4]7,22,32,33], the exponential smoothing, the ARMA [14,28,51], and the ARIMA [6,[14][15][16][17][18]24,26,[28][29][30][50][51][52] processes. The modern time series forecasters include machine learning and deep learning algorithms such as the support vector regression [6,10,11], k-nearest neighbor [10,31], artificial neural network [1,7,33], recurrent neural network (RNN) [6,9,10,12], and LSTM [6,9,10,29,30,53] algorithms. Recently, a comparison between the statistical learning and modern approaches in either simulated or real datasets with or without missing data has been reported in the literature [1,…”

“…During the last two decades, significant research work has been reported in the literature. This work has demonstrated that demand forecasting is one of the main tools for evaluating and maintaining the market and has become the cornerstone of companies' decision-making strategies [1][2][3][4][5][6][7][8][9][10][11][12]. In practice, demand forecasting includes at least two parts, namely, production forecasting and inventory control.…”

Artificial Intelligence in Supply Chain Management: Investigation of Transfer Learning to Improve Demand Forecasting of Intermittent Time Series with Deep Learning

“…Therefore, in the subsequent work, we deliberately focus on deep learning methods in the form of artificial neural networks since these can recognize non-linear causeeffect relationships. In the context of demand forecasting intermittent time series with deep learning methods, [25] and [26] have already achieved reliable results.…”