Forecasting across time series databases using recurrent neural networks on groups of similar series

a clustering approach

Kasun Bandara, Christoph Bergmeir, Slawek Smyl

Research output: Contribution to journalArticleResearchpeer-review

Abstract

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 state-of-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, together with various clustering algorithms, such as kMeans, DBScan, Partition Around Medoids (PAM), and Snob. 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.

Original languageEnglish
Article number112896
Number of pages16
JournalExpert Systems with Applications
Volume140
DOIs
Publication statusPublished - Feb 2020

Keywords

  • Big data forecasting
  • LSTM
  • Neural networks
  • RNN
  • Time series clustering

Cite this

@article{4b9602701350471f87373d5076069f0f,
title = "Forecasting across time series databases using recurrent neural networks on groups of similar series: a clustering approach",
abstract = "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 state-of-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, together with various clustering algorithms, such as kMeans, DBScan, Partition Around Medoids (PAM), and Snob. 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.",
keywords = "Big data forecasting, LSTM, Neural networks, RNN, Time series clustering",
author = "Kasun Bandara and Christoph Bergmeir and Slawek Smyl",
year = "2020",
month = "2",
doi = "10.1016/j.eswa.2019.112896",
language = "English",
volume = "140",
journal = "Expert Systems with Applications",
issn = "0957-4174",
publisher = "Elsevier",

}

Forecasting across time series databases using recurrent neural networks on groups of similar series : a clustering approach. / Bandara, Kasun; Bergmeir, Christoph; Smyl, Slawek.

In: Expert Systems with Applications, Vol. 140, 112896, 02.2020.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Forecasting across time series databases using recurrent neural networks on groups of similar series

T2 - a clustering approach

AU - Bandara, Kasun

AU - Bergmeir, Christoph

AU - Smyl, Slawek

PY - 2020/2

Y1 - 2020/2

N2 - 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 state-of-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, together with various clustering algorithms, such as kMeans, DBScan, Partition Around Medoids (PAM), and Snob. 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.

AB - 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 state-of-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, together with various clustering algorithms, such as kMeans, DBScan, Partition Around Medoids (PAM), and Snob. 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.

KW - Big data forecasting

KW - LSTM

KW - Neural networks

KW - RNN

KW - Time series clustering

UR - http://www.scopus.com/inward/record.url?scp=85071511577&partnerID=8YFLogxK

U2 - 10.1016/j.eswa.2019.112896

DO - 10.1016/j.eswa.2019.112896

M3 - Article

VL - 140

JO - Expert Systems with Applications

JF - Expert Systems with Applications

SN - 0957-4174

M1 - 112896

ER -