Simulation of an Adaptive Model Based on AIC and BIC ARIMA Predictions

Abstract

Abstract ARIMA model forecasting algorithm is a commonly used time series forecasting algorithm, this paper first obtains a stable sequence through differential operation, and then obtains a stable sequence from the AR model, as the MA model, and even the ARIMA model. Select the appropriate model for prediction and use it for adaptive mode model design. In the field of machine learning, the complexity of the model is likely to increase, while the accuracy of the model improves, and the models with a complex structure usually cause the following overfitting problem. In order to balance the complexity and the accuracy of the model reasonably, using appropriate indicators AIC (Akaike Information Criterion), as well as BIC (Bayesian information criterion), to make the judgments, which is achieved by eliciting penalty terms in the paper, and the established ARIMA (1,1,2) model meets the requirements.