Explainable Time-Series Forecasting: A Sampling-Free SHAP Approach for Transformers

This research explores enhancing the interpretability of time-series forecasting models using SHAP values, a well-established method for explaining machine learning model predictions. The utilization of a sampling-free approach suggests potential improvements in computational efficiency and practical applicability within the context of Transformers.

• •Focuses on improving the interpretability of Transformer-based time-series forecasting.
• •Employs a sampling-free SHAP method, potentially improving efficiency.
• •Targets practical application in forecasting tasks.