Decision List Prediction Using Top K Rules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Decision lists, while providing explainability in AI predictions, have inferior prediction performance compared to black box models like deep neural networks and random forests.
Innovation Solution
An information processing apparatus that calculates prediction results using the top-ranked decision rules from a decision rule set and determines the optimal decision list based on these predictions, improving prediction performance by considering multiple rules rather than just the topmost rule.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a decision list uses only the topmost rule for prediction, then explainability is maintained, but prediction performance deteriorates compared to black box models
Solution Approach 1:
The patent segments the prediction process by introducing a parameter K to select the top K ranked decision rules instead of using only the topmost rule. This segmentation allows the system to divide the contribution of multiple rules while maintaining the decision list structure, thereby improving prediction performance without completely abandoning the explainable format.
Solution Approach 2:
The patent changes the parameter from K=1 (single topmost rule) to K≥2 (multiple top-ranked rules). This parameter change enables the system to incorporate more decision rules into the prediction process, improving accuracy while still preserving the interpretability advantage of decision lists over black box models.
2Reliability
If multiple decision rules are used to improve prediction performance, then reliability improves, but the simplicity and ease of understanding deteriorates
Solution Approach 1:
The patent applies partial action by selecting only the top K ranked rules rather than using all available rules. This partial selection strikes a balance between utilizing enough rules to improve prediction performance and limiting the number to maintain ease of understanding and interpretation.
3Ease of operation
If only the topmost decision rule is applied, then ease of operation is maintained, but prediction performance worsens compared to black box models
Solution Approach 1:
The patent introduces dynamics by making the number of rules K adjustable rather than fixed at 1. This dynamic parameter allows the system to adapt between using fewer rules for better interpretability and more rules for improved performance, providing flexibility to balance ease of operation and prediction performance based on specific needs.
Data Source
AI summary
In order to improve prediction performance in prediction carried out with use of a decision list, an information processing apparatus (1) includes: a prediction section (11) that calculates a prediction result with use of predicted values of, among decision rules included in the decision list, K (K is a natural number of not less than 2) top-ranked decision rules whose conditions are satisfied by one of training examples included in a training example set; and a list determining section (12) that determines, on the basis of the prediction result given by the prediction section (11), a decision list to be output.