Machine Learning Interpretation via Feature Correlation Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current local interpretable techniques for machine learning models ignore feature correlations, making it difficult to understand and trust prediction results, especially in critical fields like finance and security where interpretability is crucial.
Innovation Solution
An information processing apparatus and method that extracts rules from a machine learning model using a decision tree method, clusters samples to identify feature correlations, and screens these rules based on data distribution to provide interpretable results for individual samples, incorporating a screening unit and interpretation unit to analyze feature correlations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature-based interpretable techniques (LIME, SHAP) are used to interpret machine learning model predictions, then local interpretability for individual samples is improved, but the correlation between features is ignored which reduces the accuracy and usefulness of interpretation
Solution Approach 1:
The patent segments the interpretation process into two distinct stages: first extracting global rules from the training data that capture feature correlations, then using these rules to guide local interpretation for individual samples. This segmentation allows the system to leverage both global patterns and local specificity, resolving the contradiction between interpretation accuracy and preserving feature correlation information.
Solution Approach 2:
The patent performs preliminary extraction of correlation-aware rules from the training dataset before conducting local interpretation. By pre-processing the data to establish rules that encode feature correlations, the system prepares the interpretative framework in advance, enabling more accurate local interpretations that respect feature relationships without requiring complex real-time analysis.
2Measurement precision
If black box machine learning models are used to achieve high prediction accuracy, then prediction performance is improved, but interpretability deteriorates making it difficult to understand and trust the results
Solution Approach 1:
The patent introduces an intermediary rule extraction layer that sits between the black box machine learning model and the user. This intermediary component translates the model's complex internal decision-making into human-readable rules that preserve the model's high accuracy while providing interpretable explanations. The rules act as a mediator that maintains the predictive power of black box models while making their behavior understandable to users.
3Ease of operation
If global interpretable techniques are used to interpret machine learning models, then overall model behavior is understood, but local accuracy deteriorates because local data distribution differs from global distribution
Solution Approach 1:
The patent applies local quality by making the interpretation process adaptive to local data characteristics. Instead of applying a single global interpretation framework uniformly, the system uses the extracted rules to tailor interpretations to local sample characteristics. This allows the interpretation method to maintain overall model understanding while adapting to local data distribution variations, thereby improving local interpretation accuracy.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The present disclosure relates to an information processing apparatus and an information processing method for interpreting processing results of a machine learning model. The information processing apparatus comprises: an extraction unit configured to extract, based on the machine learning model and a training sample set of the machine learning model, rules for processing results of the training sample set; a screening unit configured to screen, based on the distribution of the training sample set and the distribution of the rules extracted by the extraction unit, the rules extracted by the extraction unit; and an interpretation unit configured to interpret, by using the rules screened by the screening unit, a processing result of the machine learning model on a single sample to be tested. According to the information processing technique of the present disclosure, it is possible to interpret the processing results of the machine learning model for different samples based on the correlation between features of the samples.