Combining Multiple Time Series Models Through A Robust Weighted Mechanism
This work addresses the need for more accurate forecasting in time series analysis, though it is incremental as it builds on existing ensemble methods.
The paper tackles the problem of improving time series forecasting accuracy by proposing a robust weighted nonlinear ensemble technique that considers correlations among models, tested on three real-world time series, resulting in significantly lower forecast errors compared to individual models and linear combination methods.
Improvement of time series forecasting accuracy through combining multiple models is an important as well as a dynamic area of research. As a result, various forecasts combination methods have been developed in literature. However, most of them are based on simple linear ensemble strategies and hence ignore the possible relationships between two or more participating models. In this paper, we propose a robust weighted nonlinear ensemble technique which considers the individual forecasts from different models as well as the correlations among them while combining. The proposed ensemble is constructed using three well-known forecasting models and is tested for three real-world time series. A comparison is made among the proposed scheme and three other widely used linear combination methods, in terms of the obtained forecast errors. This comparison shows that our ensemble scheme provides significantly lower forecast errors than each individual model as well as each of the four linear combination methods.