Exposing Vulnerabilities in Explanation for Time Series Classifiers via Dual-Target Attacks

Interpretable time series deep learning systems are often assessed by checking temporal consistency on explanations, implicitly treating this as evidence of robustness. We show that this assumption can fail: Predictions and explanations can be adversarially decoupled, enabling targeted misclassification while the explanation remains plausible and consistent with a chosen reference rationale. We propose TSEF (Time Series Explanation Fooler), a dual-target attack that jointly manipulates the classifier and explainer outputs. In contrast to single-objective misclassification attacks that disrupt explanation and spread attribution mass broadly, TSEF achieves targeted prediction changes while keeping explanations consistent with the reference. Across multiple datasets and explainer backbones, our results consistently reveal that explanation stability is a misleading proxy for decision robustness and motivate coupling-aware robustness evaluations for trustworthy time series tasks.

Machine learning models are often used to classify time-series data such as heart signals, sensor readings, or activity records. These systems may also produce explanations that highlight which parts of the signal influenced the decision. This paper shows that such explanations can be misleading under carefully designed attacks: a model can be pushed to make a wrong target prediction while still producing an explanation that looks stable and plausible. We introduce a method called TSEF to study this risk by jointly manipulating both the prediction and the explanation. Our results suggest that simply checking whether explanations look consistent is not enough, and that future trustworthy time-series systems should evaluate the prediction and explanation together.