Explainable AI based System for Supply Air Temperature Forecast

Explainability improves trust in automated systems: The paper demonstrates that adding explanation methods (Shapley values) to a predictive model for air temperature control provides "objective justifications for curve changes," helping clients understand why the system makes specific decisions rather than just accepting predictions as a "black box."

Mathematical rigor matters for AI transparency: The authors specifically chose Shapley values over other explanation methods because they have "solid mathematical background, resulting in interpretation transparency," suggesting that responsible AI implementation should prioritize explanation techniques with formal theoretical foundations rather than ad-hoc approaches.

Feature contribution analysis enables accountability: By using Shapley values to "reveal the reasoning and highlight the contribution of each feature to the final ASAT forecast," the system allows stakeholders to identify which factors drive each decision, making it possible to audit whether the model is relying on appropriate inputs.

Context-specific explanations enhance usability: The paper's use of "contrastive explanations--slices, for each control value" shows that effective AI transparency requires tailoring explanations to specific scenarios rather than providing generic model documentation, making the system more actionable for operators.