Symmetric Decision Tree Ensemble for Feature Influence Quantification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex data science models struggle to provide clear insights into how specific features influence their outputs, making it difficult to interpret and act upon the insights derived.
Innovation Solution
A method is developed to efficiently quantify the influence of specific features on the output of a trained data science model by using an ensemble of symmetric decision trees, where each tree receives a subset of features and internal nodes at the same level designate the same splitting criterion based on the same feature.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex data science models are used to derive insights, then prediction accuracy is improved, but interpretability of feature influence deteriorates
Solution Approach 1:
The patent segments the complex data science model into an ensemble of symmetric decision trees, where each tree processes a subset of features. This segmentation allows the complex prediction task to be divided into interpretable components, enabling both accurate predictions and clear understanding of feature influence through individual tree paths and contribution values.
Solution Approach 2:
The patent introduces contribution values as a quantitative metric that 'colors' or highlights the influence of each feature on the model output. By computing and displaying these contribution values, the system makes the abstract concept of feature influence visible and measurable, transforming the black-box model into an interpretable system that shows which features pushed the prediction in which direction.
2Device complexity
If traditional methods are used to analyze feature influence, then model complexity is reduced, but computation speed deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing the structure of symmetric decision trees with their splitting criteria and feature subsets during model training. This preliminary structuring enables rapid inference during deployment, as the computation of contribution values follows predetermined paths without requiring complex real-time analysis, thus maintaining both simplicity and speed.
3Loss of information
If detailed feature contribution analysis is performed, then interpretability is improved, but computational overhead increases
Solution Approach 1:
The patent segments the feature space across multiple symmetric decision trees, where each tree handles a subset of features. This segmentation distributes the computational workload and enables parallel processing, reducing the overall computational overhead while maintaining detailed interpretability through the aggregation of contribution values from all trees.
Solution Approach 2:
The patent changes the parameter representation by using symmetric decision trees with standardized splitting criteria at each level. This parameterization allows for efficient computation of contribution values through simple path traversal and probability calculations, rather than requiring complex gradient-based or perturbation-based methods, thus reducing computational overhead while preserving interpretability.
Data Source
AI summary
A computing platform is configured to (i) receive a request to compute a score for an input data record that comprises a group of actual parameters that map to a set of features that a trained data science model is configured to receive as input; (ii) input the group into the trained data science model that comprises an ensemble of symmetric decision trees; (iii) for each individual decision tree in the ensemble: (a) identify a respective leaf such that the actual parameters satisfy splitting conditions for edges that connect nodes in a respective path from a root of the individual decision tree to the respective leaf, and (b) determine a set of respective individual contribution values for the respective leaf; and (iv) for each individual feature in the set of features, compute a respective overall contribution value based on a sum of the respective individual contribution values.