Exploiting big data in time series forecasting: A cross-sectional approach

Hartmann, Claudio; Hahmann, Martin; Lehner, Wolfgang; Rosenthal, Frank

doi:10.1109/dsaa.2015.7344786

Cited by 15 publications

(5 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To exploit the similarities between related time series, methods to build global models across sets of time series have been introduced. For example, Hartmann et al (2015) introduce a cross-sectional regression model to sets of related time series observed at the same period of time to alleviate the presence of missing values in a single time series. Also, Trapero et al (2015) use a pooled regression model by aggregating sets of related time series to produce reliable promotional forecasts in the absence of historical sales data.…”

Section: Introductionmentioning

confidence: 99%

Forecasting across time series databases using recurrent neural networks on groups of similar series: A clustering approach

Bandara

Bergmeir

Smyl

2020

Expert Systems with Applications

290

162

View full text Add to dashboard Cite

With the advent of Big Data, nowadays in many applications databases containing large quantities of similar time series are available. Forecasting time series in these domains with traditional univariate forecasting procedures leaves great potentials for producing accurate forecasts untapped. Recurrent neural networks (RNNs), and in particular Long Short Term Memory (LSTM) networks, have proven recently that they are able to outperform stateof-the-art univariate time series forecasting methods in this context, when trained across all available time series. However, if the time series database is heterogeneous, accuracy may degenerate, so that on the way towards fully automatic forecasting methods in this space, a notion of similarity between the time series needs to be built into the methods. To this end, we present a prediction model that can be used with different types of RNN models on subgroups of similar time series, which are identified by time series clustering techniques. We assess our proposed methodology using LSTM networks, a widely popular RNN variant. Our method achieves competitive results on benchmarking datasets under competition evaluation procedures. In particular, in terms of mean sMAPE accuracy it consistently outperforms the baseline LSTM model, and outperforms all other methods on the CIF2016 forecasting competition dataset.

show abstract

Section: Introductionmentioning

confidence: 99%

Forecasting across time series databases using recurrent neural networks on groups of similar series: A clustering approach

Bandara

Bergmeir

Smyl

2020

Expert Systems with Applications

290

162

View full text Add to dashboard Cite

show abstract

“…But traditional NN or SVM models cannot yet fit big data efficiently, and have therefore only been applied on a small scale (Gür Ali and Yaman, 2013;Bandara et al, 2019). With the rapid development of machine learning algorithms, such as random forest, gradient boosting trees and deep learning neural networks, recent years have seen their application to many large-scale time series forecasting problems in areas, such as transportation (Yao et al, 2018), environment (Li et al, 2016), product demand forecasting and electricity prices (Hartmann et al, 2015;Lago et al, 2018). The results, have shown the superior forecasting power of complex models with pooled data.…”

Section: Forecasting Methods For Many Time Seriesmentioning

confidence: 99%

Forecasting third-party mobile payments with implications for customer flow prediction

Fildes

2020

International Journal of Forecasting

View full text Add to dashboard Cite

“…Managers concurred that these effects are likely to be present in our case because item demands all originate from the same EMS client. The cross-learning approach has also been shown to outperform traditional univariate methods in finance (Wu et al, 2021), automotive (Gonçalves, Cortez, Carvalho, & Frazao, 2021), and retail (Spiliotis, Makridakis, Semenoglou, & Assimakopoulos, 2021) forecasting studies, in addition to well accommodating missing values and limited observations (Hartmann, Hahmann, Lehner, & Rosenthal, 2015). For multiple longitudinal series prediction tasks, cross-learning approaches are increasingly popular and proven effective in many forecasting competitions (Bojer & Meldgaard, 2021).…”

Section: Problem Reframing-data Aggregationmentioning

confidence: 99%

“…Theoretically, by exploring between-as well as within-item variation, fitted models can improve prediction performance by obtaining parameter estimates with high efficiency (using more samples) while accounting for individual heterogeneity (Greene, 2011). In addition to leveraging information across individuals, cross-item learning addresses that individual time series modeling is unable to capture common patterns and vulnerable to missing values and limited observations (Hartmann et al, 2015). Furthermore, cross-item learning affords the flexibility to include additional features, whether time variant or not, at the macro (e.g., plant and quarter fixed effects) or micro level (e.g., item-specific and block-varying attributes).…”

Section: Learn From Usementioning

confidence: 99%

Cross‐item learning for volatile demand forecasting: An intervention with predictive analytics

Chuang

Chou

Oliva

2021

J of Ops Management

View full text Add to dashboard Cite

Despite its importance to OM, demand forecasting has been perceived as a "problem-solving" exercise; most empirical work in the field has focused on explanatory models but neglected prediction problems that are part of empirical science. The present study, involving one of the leading electronics distributors in the world, aims to improve prediction accuracy under high demand volatility for procurement managers to make better inventory decisions. In response to requests for an integrated forecasting methodology, we undertook an iterative process based on three guiding principles-data pooling, theoryinformed feature engineering, and ensemble-based machine learning. The resulting framework managed to improve forecast accuracy significantly and is applicable to a broad range of situations. We present reflections and insights derived abductively through engagement with managers in this problem situation. This "problem-driven" process corresponds to intervention as a research strategy that can foster theoretical and methodological innovations in OM. Our contribution goes beyond the development of the prediction framework as it elucidates ways OM researchers could leverage theoretical foundations to inform feature derivation and model construction. We posit that this work points to a way forward to the combination of OM principles with the emerging innovations in data science and artificial intelligence.

show abstract

Exploiting big data in time series forecasting: A cross-sectional approach

Cited by 15 publications

References 7 publications

Forecasting across time series databases using recurrent neural networks on groups of similar series: A clustering approach

Forecasting across time series databases using recurrent neural networks on groups of similar series: A clustering approach

Forecasting third-party mobile payments with implications for customer flow prediction

Cross‐item learning for volatile demand forecasting: An intervention with predictive analytics

Contact Info

Product

Resources

About