Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CGS-Mask: Making Time Series Predictions Intuitive for All

Feng Lu, Wei Li, Yifei Sun, Cheng Song, Yufei Ren, Albert Y. Zomaya

TL;DR

CGS-Mask introduces a post-hoc, model-agnostic saliency method for time-series predictions that uses binary strip masks to capture temporally contiguous feature importance. By optimizing strip masks with a self-adaptive cellular genetic algorithm, CGS-Mask yields interpretable explanations with strong temporal coherence and a zero mask entropy, outperforming eight baselines on synthetic and real-world datasets. The approach is validated through extensive experiments and a pilot user study, demonstrating improved interpretability and usability in practice, including healthcare applications. Overall, CGS-Mask provides a scalable, intuitive tool for explaining time-series decisions, with potential to enhance trust and decision-making in time-sensitive domains.

Abstract

Artificial intelligence (AI) has immense potential in time series prediction, but most explainable tools have limited capabilities in providing a systematic understanding of important features over time. These tools typically rely on evaluating a single time point, overlook the time ordering of inputs, and neglect the time-sensitive nature of time series applications. These factors make it difficult for users, particularly those without domain knowledge, to comprehend AI model decisions and obtain meaningful explanations. We propose CGS-Mask, a post-hoc and model-agnostic cellular genetic strip mask-based saliency approach to address these challenges. CGS-Mask uses consecutive time steps as a cohesive entity to evaluate the impact of features on the final prediction, providing binary and sustained feature importance scores over time. Our algorithm optimizes the mask population iteratively to obtain the optimal mask in a reasonable time. We evaluated CGS-Mask on synthetic and real-world datasets, and it outperformed state-of-the-art methods in elucidating the importance of features over time. According to our pilot user study via a questionnaire survey, CGS-Mask is the most effective approach in presenting easily understandable time series prediction results, enabling users to comprehend the decision-making process of AI models with ease.

CGS-Mask: Making Time Series Predictions Intuitive for All

TL;DR

CGS-Mask introduces a post-hoc, model-agnostic saliency method for time-series predictions that uses binary strip masks to capture temporally contiguous feature importance. By optimizing strip masks with a self-adaptive cellular genetic algorithm, CGS-Mask yields interpretable explanations with strong temporal coherence and a zero mask entropy, outperforming eight baselines on synthetic and real-world datasets. The approach is validated through extensive experiments and a pilot user study, demonstrating improved interpretability and usability in practice, including healthcare applications. Overall, CGS-Mask provides a scalable, intuitive tool for explaining time-series decisions, with potential to enhance trust and decision-making in time-sensitive domains.

Abstract

Artificial intelligence (AI) has immense potential in time series prediction, but most explainable tools have limited capabilities in providing a systematic understanding of important features over time. These tools typically rely on evaluating a single time point, overlook the time ordering of inputs, and neglect the time-sensitive nature of time series applications. These factors make it difficult for users, particularly those without domain knowledge, to comprehend AI model decisions and obtain meaningful explanations. We propose CGS-Mask, a post-hoc and model-agnostic cellular genetic strip mask-based saliency approach to address these challenges. CGS-Mask uses consecutive time steps as a cohesive entity to evaluate the impact of features on the final prediction, providing binary and sustained feature importance scores over time. Our algorithm optimizes the mask population iteratively to obtain the optimal mask in a reasonable time. We evaluated CGS-Mask on synthetic and real-world datasets, and it outperformed state-of-the-art methods in elucidating the importance of features over time. According to our pilot user study via a questionnaire survey, CGS-Mask is the most effective approach in presenting easily understandable time series prediction results, enabling users to comprehend the decision-making process of AI models with ease.
Paper Structure (49 sections, 1 theorem, 20 equations, 15 figures, 4 tables, 1 algorithm)

This paper contains 49 sections, 1 theorem, 20 equations, 15 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Methodology
  4. Strip Mask
  5. Strip Mask Optimization
  6. Population Initialization
  7. Population Evolution
  8. Crossover Operator:
  9. Mutation Operator:
  10. Translation Operator:
  11. Experiments
  12. Experiments with synthetic data sets
  13. Data sets and settings
  14. Metrics
  15. Benchmarks
  16. ...and 34 more sections

Key Result

Proposition 1

If $x_{d,t}\in \mathbf{X}$ is salient for $f(\mathbf{X})$, $p_{d,t}$ is not salient for $f(\mathbf{X})$, $m_{d,t}\in \mathbf{M}$ and $m_{d,t}>0$, then $||f(\mathbf{X}) -f(\mathbf{\Gamma_{\mathbf{M}}(\mathbf{X}))}|| > 0$.

Figures (15)

  • Figure 1: Six explanation masks are employed to analyze data from MIMIC-III, covering the 48 hours preceding a patient's death. In Fig. \ref{['fig1']}(f), the green strips reveal the key features indicative of patient outcomes. Specifically, a decline in blood pressure, tachycardia of heart rate, and rapid breathing signal an impending risk of death, thereby enabling timely intervention by doctors. However, the other masks do not distinctly identify the periods and features contributing to this outcome, particularly as observed in Fig. \ref{['fig1']}(a)$-$(d).
  • Figure 2: The overall framework of CGS-Mask.
  • Figure 3: Cross entropy (CE) of saliency methods used in MIMIC-III
  • Figure 4: The user preferences for different saliency masks
  • Figure 5: User response time and selection results.
  • ...and 10 more figures

Theorems & Definitions (6)

  • Definition 1: Mask $\mathbf{M}$
  • Definition 2: Perturbation Operator $\Gamma_\mathbf{M}$
  • Proposition 1
  • proof
  • Definition 3: Perturbation Error $\Delta_{p}^{\mathbf{M}}$
  • Definition 4: Strip Mask