Deep features in time series forecasting and classification

This thesis presents the notion of dynamic and static features that surpass standard handcrafted statistical features and state-of-the-art feature selection methods in time series forecasting. Based on this notion, it proposes new methods for automatic feature extraction from time series data using a data-driven notion of similarity. The thesis also explores the impact of various data augmentation techniques on the quality of the extracted features and the performance of the final models. The research advances the field of time series analysis by presenting new techniques for feature extraction and data processing that improve the accuracy of time series forecasting and classification.