PatchDecomp: Interpretable Patch-Based Time Series Forecasting

Researchers introduce PatchDecomp, a new neural network method for time series forecasting that achieves high accuracy while providing interpretable explanations. The method divides time series into patches and shows how each patch contributes to predictions, offering both quantitative and visual insights into forecasting decisions.

• →PatchDecomp combines high forecasting accuracy with interpretability by breaking time series into subsequences called patches.
• →The method enables clear attribution of each patch's contribution to final predictions, including exogenous variables.
• →Experimental results show performance comparable to recent state-of-the-art forecasting methods on benchmark datasets.
• →The model provides both quantitative influence measures and qualitative visualization of patch-wise contributions.
• →This approach addresses the common trade-off between model complexity and interpretability in neural network forecasting.