Agentic data science with the Data Agent Kit
cloud.google.com·8h ago
AI Research
arXiv CS.AI·Aug 6·5 min read
IMFACT: Counterfactual Explanations for Time Series via Intrinsic Mode Function Substitution
TL;DR
IMFACT introduces a novel framework for generating counterfactual explanations in time series classification by substituting Intrinsic Mode Functions (IMFs). It outperforms existing methods in reliability and plausibility metrics across two benchmark datasets.
✦ Why It Matters
Implement IMFACT to enhance interpretability in your time series classification models today.
Key Takeaways
- Use IMFACT for generating counterfactuals in time series data.
- Explore different IMF-selection strategies for optimal results.
- Consider cycling through multiple NUNs for better proximity.
- Evaluate model interpretability using reliability and plausibility metrics.
How It Works
IMFACT decomposes time series signals into Intrinsic Mode Functions (IMFs) using Empirical Mode Decomposition. It then substitutes selected IMFs with those from a Nearest Unlike Neighbour (NUN) to create counterfactuals that maintain the original signal's temporal structure.
This approach allows for the generation of plausible explanations for model predictions without distorting the underlying data.