TIMING: Temporality-Aware Integrated Gradients for Time Series Explanation

Hyeongwon Jang, Changhun Kim, Eunho Yang

Abstract:

Recent explainable artificial intelligence (XAI) methods for time series primarily estimate point-wise attribution magnitudes, while overlooking the directional impact on predictions, leading to suboptimal identification of significant points. Our analysis shows that conventional Integrated Gradients (IG) effectively capture critical points with both positive and negative impacts on predictions. However, current evaluation metrics fail to assess this capability, as they inadvertently cancel out opposing feature contributions. To address this limitation, we propose novel evaluation metrics—Cumulative Prediction Difference (CPD) and Cumulative Prediction Preservation
(CPP)—to systematically assess whether attribution methods accurately identify significant positive and negative points in time series XAI. Under these metrics, conventional IG outperforms recent counterparts. However, directly applying IG to time series data may lead to suboptimal outcomes, as generated paths ignore temporal relationships and introduce out-of-distribution samples. To overcome these challenges, we introduce TIMING, which enhances IG by incorporating temporal awareness while maintaining its theoretical properties. Extensive experiments on synthetic and real-world time series benchmarks demonstrate that TIMING outperforms existing time
series XAI baselines. Our code is available at https://github.com/drumpt/TIMING.

Lay Summary: Time series data—heartbeat traces, financial data, and sensor measurements—drive critical decisions in safety-critical domains. Accurate predictions are essential, but explaining model decisions is equally crucial for building trust. Most XAI algorithms report only how much each moment matters, not whether it helps or hurts predictions, causing evaluation metrics to obscure important points by canceling opposing effects. We introduce Cumulative Prediction Difference (CPD) and Cumulative Prediction Preservation (CPP) metrics that measure "helpful" and "harmful" moments without letting opposite effects blur together. Under these evaluations, classic Integrated Gradients (IG) outperforms newer alternatives. However, IG ignores temporal structure in time series data. To address this limitation, we develop TIMING, which enhances Integrated Gradients by incorporating sequence information. Across simulated and real-world datasets, TIMING identifies critical moments more accurately than existing tools. This provides clinicians, analysts, and engineers with clearer insight into AI decisions. Our code and evaluation suite are publicly available for others to utilize and build upon.

Link To Code: https://github.com/drumpt/TIMING

Primary Area: Applications->Time Series

Keywords: Time Series, XAI, Explainability

Submission Number: 15165