AI Prediction Interpretation via Feature Contribution Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users of AI systems face challenges in interpreting the basis of prediction results due to the black-box nature of machine learning models, making it difficult to trust and understand the reliability of the predictions, especially in fields like medicine and finance.
Innovation Solution
A computer system that includes a processor and memory, storing model information and interpretation factor conversion information, calculates contribution values for feature quantities, and generates display information to present interpretation factors and their scores, enabling users to understand the basis of AI predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used to improve prediction accuracy, then prediction accuracy is improved, but interpretability of the prediction basis deteriorates
Solution Approach 1:
The patent introduces an intermediary system that includes a contribution degree calculation unit and an interpretation factor generation unit. These intermediaries bridge the gap between the black-box ML model and the user by calculating contribution degrees of feature quantities and generating human-interpretable factors, thus preserving prediction accuracy while restoring interpretability through intermediate processing steps.
Solution Approach 2:
The patent segments the prediction process into distinct components: the ML model generates prediction results, while separate units calculate contribution degrees for each feature quantity and generate interpretation factors. This segmentation allows the system to maintain the accuracy benefits of complex ML models while providing transparent, breakdown explanations for each prediction through individual feature contribution analysis.
2Loss of information
If contribution degree calculation methods are used to explain feature importance, then information about feature influence is provided, but user understanding and trust in predictions deteriorate
Solution Approach 1:
The patent transforms the abstract parameter of contribution degree into concrete interpretation factors by comparing feature quantity values against reference data and generating descriptive factors. This parameter transformation converts numerical contribution degrees into human-understandable explanations that users can interpret and trust, such as describing how specific feature values influence predictions in context.
Solution Approach 2:
The interpretation factor generation unit acts as an intermediary that translates contribution degree calculations into user-friendly explanations. It compares input feature values with reference data, calculates contribution degrees, and generates interpretable factors that bridge the gap between technical model outputs and user comprehension, thereby building trust through transparent reasoning.
3Productivity
If AI prediction systems are implemented to automate decision-making, then productivity is improved, but reliability of prediction basis deteriorates
Solution Approach 1:
The patent implements feedback mechanisms where prediction results are accompanied by interpretation factors that explain the basis of each prediction. This feedback loop allows users to verify prediction reliability by examining the generated interpretation factors, which compare feature values against reference data and provide transparent reasoning, thereby maintaining trust in automated decision-making systems.
Solution Approach 2:
The system performs preliminary actions by pre-calculating reference data and preparation information before making predictions. It stores reference feature quantities and preparation data that are used to generate interpretation factors, enabling the system to provide reliable, explainable predictions in advance rather than requiring post-hoc analysis, thus enhancing both automation and reliability.
Data Source
AI summary
A computer system stores interpretation factor conversion information for managing an interpretation factor interpreting a basis of a prediction result for input data, the interpretation factor is determined by a value of each of a plurality of feature quantities contained in the input data including values of the plurality of feature quantities, and a first evaluation value of each of the plurality of feature quantities contained in the input data. When evaluation target data is input, the computer system calculates a prediction result, calculates a contribution value of each of the plurality feature quantities contained in the evaluation target data, specifies a corresponding interpretation factor, based on a value and a contribution value of each of the plurality of feature quantities contained in the evaluation target data, by referring to the interpretation factor conversion information, and generates and outputs display information for presenting the specified interpretation factor.