Pre-trained Forecasting Models: Strong Zero-Shot Feature Extractors for Time Series Classification

Recent advancements in time series forecasting models have opened up exciting possibilities for classification tasks. This study explores how pre-trained models, typically used for forecasting, can also serve as powerful feature extractors for classification. By comparing various extraction strategies and introducing innovative embedding techniques, the research demonstrates that these models can effectively enhance classification performance. This is significant as it broadens the applicability of forecasting models, potentially leading to improved outcomes in various fields that rely on time series data.