Enhancing Time Series Forecasting via a Parallel Hybridization of ARIMA and Polynomial Classifiers

Abstract

Time series forecasting is an important task in many fields such as economics, industry, and environmental studies. Many forecasting methods have been developed, including statistical methods and deep learning models. Among them, the Auto-Regressive Integrated Moving Average (ARIMA) model is widely used for modeling linear patterns in time series data. In contrast, polynomial classifiers are effective in modeling non-linear relationships and have shown good performance in several prediction problems. This paper proposes a hybrid forecasting model that combines ARIMA and a polynomial classifier in a parallel structure. The ARIMA model is used to capture linear patterns, while the polynomial classifier is used to model non-linear behavior. The outputs of the two models are combined to obtain the final prediction. Experimental results on four different datasets show that the hybrid model can provide competitive forecasting performance. For example, on the Delhi Climate dataset, the hybrid model achieved an RMSE of 1.6698, compared with 1.7055 for the corresponding standalone first-order polynomial classifier. On the gold price dataset, the hybrid model achieved an RMSE of 29.3236 when combined with the first-order polynomial classifier, compared with 30.5953 for the corresponding standalone configuration. However, the hybrid model did not outperform every polynomial-classifier configuration, as the standalone second-order polynomial classifier achieved a lower RMSE of 25.5762 on the gold price dataset. These results indicate that the effectiveness of the proposed hybrid approach depends on the dataset and the polynomial-classifier configuration considered.