Model identification for ARMA time series through convolutional neural\n networks

2018/04/11 by Wai Hoh Tang, Tang, Wai Hoh, Adrian Röllin +1
Computer Science · Decision Sciences · Engineering · #62M10 #Computation (stat.CO) #FOS: Computer and information sciences #Fault Detection and Control Systems #Machine Learning (cs.LG) #Methodology (stat.ME) #Neural Networks and Applications #Stock Market Forecasting Methods

paper · pdf · doi:10.48550/arxiv.1804.04299

openalex publication_date 2018/04/11 · openalex created_date 2022/08/07 · openalex updated_date 2026/07/28

Abstract

In this paper, we use convolutional neural networks to address the problem of\nmodel identification for autoregressive moving average time series models. We\ncompare the performance of several neural network architectures, trained on\nsimulated time series, with likelihood based methods, in particular the Akaike\nand Bayesian information criteria. We find that our neural networks can\nsignificantly outperform these likelihood based methods in terms of accuracy\nand, by orders of magnitude, in terms of speed.\n

Related

• ARMA Cell: A Modular and Effective Approach for Neural Autoregressive Modeling
• Fitting ARMA Time Series Models without Identification: A Proximal Approach
• Revisiting Inference for ARMA Models: Improved Fits and Superior Confidence Intervals
• Online Learning for Time Series Prediction
• ARMA Time-Series Modeling with Graphical Models