Feature Set Extraction for Explainable AI Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of maintaining prediction accuracy while improving explainability in AI models is exacerbated by the generation of a large number of feature sets through data mining, leading to deteriorated interpretability and increased multicollinearity, which complicates regression analysis and hampers effective feature identification.
Innovation Solution
A method involving an extraction device that selects a specific number of feature sets based on statistical information, using subset extraction and combination generation units to output combinations that enhance explainability and reduce the number of explanatory variables, thereby mitigating multicollinearity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of feature types included in prediction target data is increased to improve prediction accuracy, then prediction accuracy is improved, but it becomes difficult to identify which feature is useful for prediction, thus deteriorating the explainability of prediction
Solution Approach 1:
The patent extracts only the necessary and useful feature sets from the large number of generated feature sets by using statistical information and significance tests. This extraction process removes redundant and irrelevant feature sets, retaining only those that contribute meaningfully to prediction, thereby maintaining prediction accuracy while improving explainability.
Solution Approach 2:
The patent changes the parameter of feature set selection by introducing statistical significance thresholds and extraction criteria. Instead of using all generated feature sets, the system applies parameter-based filtering to select only those feature sets that meet predetermined statistical criteria, thus balancing prediction accuracy with interpretability.
2Adaptability or versatility
If data mining is used to generate a combination of feature types useful for making prediction, then the number of feature types is increased, but the number of generated feature sets increases, making it difficult to interpret prediction basis
Solution Approach 1:
The patent applies extraction by selecting only the most significant feature sets from the large number generated through data mining. By using statistical tests and significance criteria, the system extracts a manageable subset of feature sets that maintain predictive power while reducing complexity and improving interpretability.
Solution Approach 2:
Instead of processing all generated feature sets, the patent applies partial action by selectively processing only those feature sets that meet statistical significance thresholds. This partial processing approach avoids the complexity of handling all possible feature sets while still capturing the essential predictive information.
3Reliability
If a large number of feature sets are generated by data mining, then more comprehensive prediction coverage is achieved, but multicollinearity increases and regression analysis becomes complicated
Solution Approach 1:
The patent extracts feature sets that are statistically significant and removes those that contribute to multicollinearity. By applying extraction based on statistical criteria, the system maintains comprehensive prediction coverage while eliminating redundant and correlated feature sets that complicate regression analysis.
Solution Approach 2:
The patent changes the selection parameters for feature sets by introducing statistical significance thresholds and correlation-based filtering. This parameter change approach ensures that only feature sets with meaningful predictive power and low multicollinearity are retained, simplifying regression analysis while maintaining prediction coverage.
Data Source
AI summary
A non-transitory computer-readable recording medium has stored therein an extraction program that causes a computer to execute a process. The process includes extracting a plurality of subsets from a data set including a plurality of pieces of data including a feature quantity of each of a plurality of feature types, the plurality of subsets each including part of the plurality of pieces of data. The process includes obtaining, using each of the plurality of subsets, a combination of features useful for data prediction. The process includes extracting a specific number of combinations from a plurality of the combinations obtained from the plurality of subsets, the extracting being based on statistical information regarding each of the plurality of combinations. The process includes outputting the specific number of combinations.