Reimagining service delivery in the agentic era with Google Public Sector
cloud.google.com·21h ago
AI Research
arXiv CS.AI·Jun 29·5 min read
Global Explanations for Multivariate Time Series Forecasting Models via $K$-Order Markov Approximations
TL;DR
Multivariate time series forecasting models often lack interpretability, making it difficult to understand their predictions. A new method using $K$-Order Markov Approximations was developed to provide global explanations for these models.
✦ Why It Matters
Engineers can use $K$-Order Markov Approximations to improve the interpretability of their forecasting models.
Key Takeaways
- Utilize KARMA for better explanations of time-series models.
- Determine the optimal history length $K$ for your data.
- Leverage the five-level explanation hierarchy for insights.
- Compare KARMA's performance against traditional methods like TimeSHAP.
How It Works
KARMA constructs a Markov surrogate model that captures temporal dependencies by estimating a $K$-order Markov transition kernel. It identifies the minimal history length $K$ necessary for accurate predictions, allowing for a structured explanation hierarchy that reflects the causal relationships in the data.