Adaptive Online Learning for the Autoregressive Integrated Moving Average Models

This paper addresses the problem of predicting time series data using autoregressive integrated moving average (ARIMA) model in an online manner. Existing algorithms require selection, which is consuming and unsuitable for setting learning. Using adaptive learning techniques, we develop fitting ARIMA models without hyperparameters. The regret analysis experiments on both synthetic real-world datasets show that performance proposed can be guaranteed theory practice.