Sampling shifts challenge time series prediction robustness
Domain Generalization under Sampling Pattern Shifts in Irregular Time Series
Machine LearningArtificial Intelligence
Summary
Irregularly timed data, like medical records taken at uneven intervals, can change a lot across different situations, making predictions tricky. The authors found that changes in how data is collected can confuse existing prediction methods and cause errors. They created a new test setup to study this problem and developed PRISM, a method that learns both the features and the sampling patterns separately without knowing the outcomes first. This approach helps the model avoid relying on shortcuts that fail when the sampling changes, improving prediction reliability in new data with different timing.
What this means in practice
- •For medical data teams: Improve patient outcome prediction models by handling changes in how medical data is sampled over time.
- •For industrial iot engineers: Build more reliable predictive maintenance systems that remain effective despite varying data collection times from sensors.
Authors
Changhun Kim, Joohyung Lee, Kwanhyung Lee, Donghwee Yoon, Grigorios Chrysos, Eunho Yang
Abstract
Irregularly sampled multivariate time series (ISMTS) are prevalent in real-world applications, where both observation times and available measurements can vary substantially across domains. While recent models increasingly exploit such sampling information for prediction, its robustness under sampling pattern shifts remains underexplored. We introduce HAR-C, to the best of our knowledge the first controlled benchmark for sampling pattern shifts in ISMTS, and show that sampling shifts alone can substantially degrade performance, induce sampling-specific shortcuts, and remain challenging for existing domain generalization (DG) methods. Motivated by these findings, we propose PRISM, a DG framework that first learns complementary feature-centric and sampling-centric representations without task labels, and subsequently performs robust supervised training across diverse sampling variations to discourage brittle shortcut reliance. Extensive experiments on controlled and real-world ISMTS benchmarks demonstrate that PRISM consistently improves robustness to unseen sampling shifts over existing methods. Our code is available at https://anonymous.4open.science/r/PRISM.