Hierarchical Feature Decomposition for Interpretable Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex machine learning models, particularly those used for predictive tasks, often generate results that are difficult to interpret, making it unclear how the model's outputs relate to the task it was designed to perform, especially as tasks become more intricate.
Innovation Solution
The implementation of a computer-implemented method that uses a hierarchical structure to decompose features into partitions and sub-partitions, allowing for the computation of interaction scores between these components, thereby providing interpretability to the predictive models by assigning weights to each feature based on its contribution to the final prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex machine learning models are used to perform complicated tasks, then the model's predictive capability is improved, but the interpretability of the model results deteriorates
Solution Approach 1:
The patent segments the complex machine learning model into multiple interpretable components including a hierarchical structure of rules, decision paths, and feature groups. Each component can be individually understood and explained, while collectively they maintain the model's predictive power. The segmentation allows users to trace predictions through specific decision paths and understand the contribution of individual features.
Solution Approach 2:
The patent introduces an intermediary layer between the complex model and the user, consisting of explanation artifacts such as decision paths, rule descriptions, and feature importance scores. This intermediary translates the complex internal representations into human-understandable formats, preserving interpretability while maintaining access to complex predictive capabilities.
2Adaptability or versatility
If the model structure is made more complex to handle intricate tasks, then the task performance is improved, but the clarity of the relationship between model results and the task deteriorates
Solution Approach 1:
The complex model is segmented into modular components with clear functional definitions. Each segment handles specific aspects of the task, and the relationships between segments are made explicit through structured decision paths and rule hierarchies. This segmentation maintains adaptability while improving clarity of the model-task relationship.
Solution Approach 2:
The patent adds a new dimension of explanation to the model output, providing not just predictions but also structured interpretations along multiple dimensions: feature-level explanations, rule-level explanations, and path-level explanations. This dimensional expansion maintains task performance while clarifying the model-task relationship.
Data Source
AI summary
A computer-implemented method for providing interpretable predictions from a machine learning model includes receiving a data structure that represents a hierarchical structure of a set of features (X) used by one or more predictive models to generate a set of predictions (Y). An interpretability model is built corresponding to the predictive models, by assigning an interpretability to each prediction Yi based on the hierarchical structure. Assigning the interpretability includes decomposing X into a plurality of partitions Xj using the hierarchical structure, wherein X=U1NXj, N being the number of partitions. Further, each partition is decomposed into a plurality of sub-partitions using the hierarchical structure until atomic sub-partitions are obtained. A score is computed for each partition as a function of the predicted scores of the sub-partitions, wherein the predicted scores represent interactions between the sub-partitions. Further, an interpretation of a prediction is outputted.