Model identification for ARMA time series through convolutional neural networks

We use convolutional neural networks for model identification in ARMA time series models, where our are trained on synthetic data with known ground truths. Comparing the performance of these traditional likelihood-based methods, particular Akaike and Bayesian Information Criteria, we able to show that when it comes statistical inference orders, can significantly outperform methods terms accuracy and, by orders magnitude, speed. also observe improvements forecasting. Our approach shows feasibility using artificial situations classical difficult or costly implement.