Improving the Accuracy of Global Forecasting Models using Time Series\n Data Augmentation

2020/08/06 by Kasun Bandara, Bandara, Kasun, Hansika Hewamalage +7 · 4 citations
Computer Science · Decision Sciences · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Forecasting Techniques and Applications #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Stock Market Forecasting Methods #Time Series Analysis and Forecasting

paper · pdf · doi:10.48550/arxiv.2008.02663

openalex publication_date 2020/08/06 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28

Abstract

Forecasting models that are trained across sets of many time series, known as\nGlobal Forecasting Models (GFM), have shown recently promising results in\nforecasting competitions and real-world applications, outperforming many\nstate-of-the-art univariate forecasting techniques. In most cases, GFMs are\nimplemented using deep neural networks, and in particular Recurrent Neural\nNetworks (RNN), which require a sufficient amount of time series to estimate\ntheir numerous model parameters. However, many time series databases have only\na limited number of time series. In this study, we propose a novel, data\naugmentation based forecasting framework that is capable of improving the\nbaseline accuracy of the GFM models in less data-abundant settings. We use\nthree time series augmentation techniques: GRATIS, moving block bootstrap\n(MBB), and dynamic time warping barycentric averaging (DBA) to synthetically\ngenerate a collection of time series. The knowledge acquired from these\naugmented time series is then transferred to the original dataset using two\ndifferent approaches: the pooled approach and the transfer learning approach.\nWhen building GFMs, in the pooled approach, we train a model on the augmented\ntime series alongside the original time series dataset, whereas in the transfer\nlearning approach, we adapt a pre-trained model to the new dataset. In our\nevaluation on competition and real-world time series datasets, our proposed\nvariants can significantly improve the baseline accuracy of GFM models and\noutperform state-of-the-art univariate forecasting methods.\n

Cited by

• Data Augmentation in Time Series Forecasting through Inverted Framework
• A Review of the Long Horizon Forecasting Problem in Time Series Analysis
• The effects of retraining on the stability of global models in retail demand forecasting
• The cost of ensembling: is it always worth combining?

Related

• DAM: Towards A Foundation Model for Time Series Forecasting
• Time Series Data Augmentation as an Imbalanced Learning Problem
• Towards a General Time Series Forecasting Model with Unified Representation and Adaptive Transfer
• N-BEATS: Neural basis expansion analysis for interpretable time series forecasting
• Do We Really Need Deep Learning Models for Time Series Forecasting?