AIE-values for Interpretable SHAP Metrics in Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning forecasting models generate raw SHAP quantities that are difficult for planners to interpret, especially in demand or supply planning contexts, due to their complex mathematical nature and requirement for specific skills, leading to potential misinterpretation and incorrect conclusions.
Innovation Solution
The system generates Applied and Indicative Explanation values (AIE-values) based on conventional SHAP quantities from existing machine learning forecasting engines, providing relative magnitude and direction effects across forecast items and times, allowing for groupings by scale and normalized values, which are then displayed in a graphical user interface for improved usability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw SHAP quantities are provided directly from machine learning models, then measurement precision of feature contributions is improved, but ease of operation for non-data science experts deteriorates
Solution Approach 1:
The patent introduces an intermediary processing layer that transforms raw SHAP quantities into enhanced metrics (AIE-values). This mediator component performs normalization, scaling, and aggregation operations to make the data interpretable for planners while preserving the underlying mathematical precision of feature contributions.
Solution Approach 2:
The system applies parameter transformations to SHAP quantities including normalization to standard scales, aggregation across multiple features, and conversion into standardized metrics. These parameter changes maintain the mathematical relationships while presenting data in a form that is intuitive for non-expert users.
2Measurement precision
If detailed SHAP quantities for each feature are provided, then measurement precision of individual feature effects is improved, but device complexity increases
Solution Approach 1:
The patent segments the complex SHAP analysis into distinct processing stages: individual feature contribution calculation, normalization of each feature's SHAP values, aggregation across features, and final presentation formatting. This segmentation reduces overall system complexity by breaking down the complex transformation process into manageable, modular operations.
Solution Approach 2:
The system systematically transforms raw SHAP parameters through standardized operations including normalization to common scales, aggregation functions, and conversion to interpretable metrics. These parameter changes simplify the data structure while preserving essential information about feature contributions.
3Ease of operation
If normalized and aggregated SHAP metrics are generated, then ease of operation for decision-making is improved, but loss of information about individual feature contributions occurs
Solution Approach 1:
The patent implements a nested structure where aggregated metrics contain within them the underlying individual feature contributions. The enhanced metrics are constructed to preserve hierarchical relationships, allowing users to drill down from aggregated views to individual feature-level details without losing the original SHAP quantity information.
Solution Approach 2:
The system creates multi-functional metrics that serve multiple purposes simultaneously: they provide normalized scales for comparison across different features, maintain aggregated summaries for high-level decision-making, and preserve individual feature contribution data for detailed analysis. This universal metric structure eliminates the need to choose between aggregation and detail.
Data Source
AI summary
A system and method for generating a graphical user interface including enhanced metrics based on the SHAP quantities of a machine learning forecast model may include: receiving forecast data related to a machine learning forecast model that includes, for each input category of each forecast item at each forecast time, a SHAP quantity and a sign indicating a directionality of the SHAP quantity, based on the SHAP quantities, generating relative magnitude values, directional significance values, and relative contribution values, and outputting in a graphical user interface the relative magnitude values, the directional significance values, and the relative contribution values as Applied and Indicative Explanation values (AIE-values).