Optimizing accuracy and diversity: a multi-task approach to forecast combinations
Researchers have developed a novel multi-task deep learning approach for time series forecasting that optimizes both accuracy and diversity in model combinations. This method jointly selects and combines forecasting models by extracting features to determine optimal weights and identify diverse, accurate prediction methods for each time series. Experiments on the M4 competition dataset and real-world traffic data demonstrate improved point forecast accuracy over existing state-of-the-art techniques. AI
IMPACT Introduces a novel deep learning architecture for improving time series forecasting accuracy and diversity.