Time Series Extrinsic Regression

Time series extrinsic regression (TSER) focuses on predicting a continuous outcome variable using a set of time series as predictors, where the outcome is not directly embedded within the time series themselves. Recent research emphasizes developing robust algorithms capable of handling high-dimensional, variable-length, and potentially incomplete time series data, with a focus on ensemble methods like rotation forests and novel architectures such as tree-based ensembles (e.g., DrCIF) and feature-based pipelines (e.g., FreshPRINCE) that leverage summary statistics and convolutional approaches. These advancements improve predictive accuracy and offer interpretability through feature-based explanations, impacting diverse fields like human activity recognition and earth observation where time series data is prevalent.