Shapelets-Enriched Selective Forecasting using Time Series Foundation Models

TL;DR

The paper tackles the reliability gap of time-series foundation models in critical segments by introducing a shapelet-guided selective forecasting framework. It learns shift-invariant shapelets from high-error validation data and uses z-normalized distance to selectively discard unreliable test predictions, while leveraging both zero-shot and full-shot fine-tuned TSFMs (notably Tiny Time Mixer). The approach yields substantial MSE reductions (averages of $22.17\%$ for zero-shot and $22.62\%$ for full-shot) and outperforms random baselines on several datasets, while providing interpretable insight into where and why predictions are discarded. This work enhances practical forecasting reliability for TSFMs with modest computational overhead and opens avenues for adaptive, data-driven pattern-based rejection in time-series tasks.

Abstract

Time series foundation models have recently gained a lot of attention due to their ability to model complex time series data encompassing different domains including traffic, energy, and weather. Although they exhibit strong average zero-shot performance on forecasting tasks, their predictions on certain critical regions of the data are not always reliable, limiting their usability in real-world applications, especially when data exhibits unique trends. In this paper, we propose a selective forecasting framework to identify these critical segments of time series using shapelets. We learn shapelets using shift-invariant dictionary learning on the validation split of the target domain dataset. Utilizing distance-based similarity to these shapelets, we facilitate the user to selectively discard unreliable predictions and be informed of the model's realistic capabilities. Empirical results on diverse benchmark time series datasets demonstrate that our approach leveraging both zero-shot and full-shot fine-tuned models reduces the overall error by an average of 22.17% for zero-shot and 22.62% for full-shot fine-tuned model. Furthermore, our approach using zero-shot and full-shot fine-tuned models, also outperforms its random selection counterparts by up to 21.41% and 21.43% on one of the datasets.

Figures (14)

• Figure 1: Workflow diagram showing the selective prediction framework guided by shapelets.
• Figure 2: Performance comparison of Selective Forecasting using shapelets against baselines.
• Figure 3: The plot showing a subset of shapelets learned from selected validation samples of Exchange Rate.
• Figure 4: A test sample selected for discarding predictions with the best matching shapelet for Exchange Rate.
• Figure 5: A test sample selected for making predictions with the best matching shapelet for Exchange Rate.