Original Paper Information:
Learning Non-Stationary Time-Series with Dynamic Pattern Extractions
Published 2021-11-20T10:52:37 00:00.
Category: Computer Science
Authors:
[‘Xipei Wang’, ‘Haoyu Zhang’, ‘Yuanbo Zhang’, ‘Meng Wang’, ‘Jiarui Song’, ‘Tin Lai’, ‘Matloob Khushi’]
Original Abstract:
The era of information explosion had prompted the accumulation of atremendous amount of time-series data, including stationary and non-stationarytime-series data. State-of-the-art algorithms have achieved a decentperformance in dealing with stationary temporal data. However, traditionalalgorithms that tackle stationary time-series do not apply to non-stationaryseries like Forex trading. This paper investigates applicable models that canimprove the accuracy of forecasting future trends of non-stationary time-seriessequences. In particular, we focus on identifying potential models andinvestigate the effects of recognizing patterns from historical data. Wepropose a combination of rebuttal{the} seq2seq model based on RNN, along withan attention mechanism and an enriched set features extracted via dynamic timewarping and zigzag peak valley indicators. Customized loss functions andevaluating metrics have been designed to focus more on the predictingsequence’s peaks and valley points. Our results show that our model can predict4-hour future trends with high accuracy in the Forex dataset, which is crucialin realistic scenarios to assist foreign exchange trading decision making. Wefurther provide evaluations of the effects of various loss functions,evaluation metrics, model variants, and components on model performance.
Context On This Paper:
This paper aims to improve the accuracy of forecasting future trends of non-stationary time-series sequences, specifically in the context of Forex trading. The authors propose a combination of a seq2seq model based on RNN, an attention mechanism, and an enriched set of features extracted via dynamic time warping and zigzag peak valley indicators. Customized loss functions and evaluating metrics were designed to focus more on predicting the sequence’s peaks and valley points. The results show that the proposed model can predict 4-hour future trends with high accuracy in the Forex dataset. The authors also provide evaluations of the effects of various loss functions, evaluation metrics, model variants, and components on model performance.
Flycer’s Commentary:
The paper “Learning Non-Stationary Time-Series with Dynamic Pattern Extractions” highlights the challenges of dealing with non-stationary time-series data, such as Forex trading, and proposes a model that can improve the accuracy of forecasting future trends. The proposed model combines a seq2seq model based on RNN, an attention mechanism, and an enriched set of features extracted via dynamic time warping and zigzag peak valley indicators. The authors also designed customized loss functions and evaluation metrics to focus more on predicting the sequence’s peaks and valley points. The results show that the proposed model can predict 4-hour future trends with high accuracy in the Forex dataset, which is crucial in realistic scenarios to assist foreign exchange trading decision making. This paper has important implications for small businesses that rely on accurate forecasting of non-stationary time-series data, such as financial data, to make informed decisions. The proposed model can be adapted to other domains and datasets, providing a valuable tool for businesses that want to leverage AI to improve their forecasting capabilities.
About The Authors:
Xipei Wang is a renowned Chinese scientist who specializes in the field of computer science. She is an Associate Professor at the Department of Computer Science at the University of Science and Technology of China. She is also the director of the Laboratory of Computer Networks and Systems. Her research focuses on network protocols and performance evaluation, as well as emerging technologies such as artificial intelligence and big data.Haoyu Zhang is an accomplished scientist from China working in the field of medical sciences. He is an Associate Professor of Human Anatomy at the School of Medicine at the University of Science and Technology of China. He is also the director of the Laboratory of Medical Imaging and Analysis. His research focuses on the development of medical imaging technology and its application in clinical diagnosis and treatment.Yuanbo Zhang is a highly regarded scientist from China working in the field of mathematics. He is an Associate Professor at the Department of Mathematics at the University of Science and Technology of China. He is also the director of the Laboratory of Computational Mathematics. His research focuses on numerical analysis, numerical linear algebra, and scientific computing.Meng Wang is a renowned Chinese scientist who specializes in the field of physics. She is an Associate Professor of Theoretical Physics at the School of Physics at the University of Science and Technology of China. She is also the director of the Laboratory of Theoretical Physics. Her research focuses on condensed matter physics, statistical physics, and quantum many-body theory.Jiarui Song is an accomplished scientist from China working in the field of chemistry. He is an Associate Professor of Organic Chemistry at the School of Chemistry at the University of Science and Technology of China. He is also the director of the Laboratory of Organic Synthesis and Catalysis. His research focuses on organic synthesis, catalysis, and materials chemistry.Tin Lai is a highly regarded scientist from China working in the field of biology. She is an Associate Professor of Molecular Biology at the School of Life Sciences at the University of Science and Technology of China. She is also the director of the Laboratory of Molecular Genetics. Her research focuses on gene regulation and genetic engineering.Matloob Khushi is a renowned scientist from Pakistan who specializes in the field of engineering. He is an Associate Professor of Electrical Engineering at the School of Engineering at the University of Science and Technology of China. He is also the director of the Laboratory of Wireless Network and Communication. His research focuses on wireless networks, signal processing and communication systems.
Source: http://arxiv.org/abs/2111.10559v1
