Decision Tree Model Approximation for Interpretable Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, particularly those used in prediction tasks, are difficult for users to interpret due to their complex structure and lack of transparency, making it challenging for business experts and others to understand how features interact and contribute to predictions, which hinders their practical application.
Innovation Solution
The method involves approximating complex machine learning models, such as deep learning models, into decision tree models, allowing for visual presentation and interpretation. This is achieved by acquiring prediction results, generating decision tree training samples, training a decision tree model, and visually presenting it, enabling users to understand the prediction mechanism.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used for prediction tasks, then prediction accuracy is improved, but model interpretability deteriorates
Solution Approach 1:
The patent creates a simplified copy (decision tree model) that replicates the prediction behavior of the complex machine learning model. This copy maintains the predictive functionality while being visually interpretable, allowing users to understand the decision-making process without sacrificing prediction accuracy
Solution Approach 2:
The patent introduces an intermediary visualization layer that mediates between the complex machine learning model and the user. This visualization layer translates the internal workings of the model into an understandable format, bridging the gap between accuracy and interpretability
2Productivity
If complex machine learning models are used, then prediction capability is improved, but understanding of feature interaction deteriorates
Solution Approach 1:
The patent segments the complex model's decision-making process into discrete, visualizable components. By breaking down the prediction process into separate decision nodes and branches, users can understand how different features interact at each stage without being overwhelmed by the overall complexity
3Measurement precision
If deep learning models with multiple layers are used, then prediction performance is improved, but model complexity increases
Solution Approach 1:
The patent creates a simplified decision tree copy that replicates the prediction behavior of the complex deep learning model. This copy maintains the essential predictive functionality while eliminating the unnecessary complexity of multiple hidden layers and neurons, making the model easier to understand and interpret
Data Source
AI summary
A method and device for presenting a prediction model, and a method and device for adjusting a prediction model. The method for presenting a prediction model includes: obtaining at least one prediction result of a prediction model for at least one prediction sample; obtaining at least one decision-making tree training sample for training a decision-making tree model according to the at least one prediction sample and the at least one prediction result, the decision-making tree model being used for fitting the prediction model; training the decision-making tree model by using at least one decision-making tree training sample; and visually presenting the trained decision-making tree model. By means of the method, a prediction model hard to understand can be approximated to a decision-making tree model, and the approximated decision-making tree model is presented, so that a user better understands the prediction model according to the presented decision-making tree model.