Shapley Value Allocation for Engineered Feature Interpretation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models lack effective methods for interpreting feature importance and explaining predictions, particularly in complex models with engineered features and interactions, leading to difficulties in understanding and improving model performance.
Innovation Solution
The system calculates Shapley values for both original and engineered features, using feature lineage to allocate values back to their parent features, and employs heuristics to detect interaction effects, enabling more efficient computation and visualization of feature contributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex machine learning models with engineered features are used to improve prediction accuracy, then model accuracy is improved, but model interpretability deteriorates
Solution Approach 1:
The patent segments the interpretation problem into local explanations (individual predictions) and global explanations (overall model behavior). It further segments the feature attribution into engineered features and their parent features, allowing hierarchical analysis that maintains interpretability while working with complex models.
Solution Approach 2:
The patent introduces Shapley values as an intermediary mechanism to bridge the gap between complex model predictions and human-understandable feature contributions. Shapley values serve as a mediator that quantifies the marginal contribution of each feature, making the black-box model's decision-making process interpretable.
2Loss of information
If Shapley values are calculated for all features including engineered features, then complete feature attribution is achieved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-identifying feature lineage relationships and parent-feature mappings before calculating Shapley values. This allows the system to efficiently aggregate results and avoid redundant computations, reducing the overall computational burden while maintaining complete attribution.
Solution Approach 2:
The patent merges the calculation of Shapley values for engineered features with their parent features through aggregation. By combining results hierarchically and utilizing the linearity property of Shapley values, the system reduces the number of separate calculations needed while preserving complete feature attribution.
Data Source
AI summary
Systems and methods are described for interpreting machine learning model predictions. An example method includes: providing a machine learning model configured to receive a plurality of features as input and provide a prediction as output, wherein the plurality of features includes an engineered feature including a combination of two or more parent features; calculating a Shapley value for each feature in the plurality of features; and allocating a respective portion of the Shapley value for the engineered feature to each of the two or more parent features.