Post-modeling Category Merging for Predictive Model Simplification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods lack the capability to deeply analyze the nuances of categorical variables, leading to inefficiencies in feature evaluation and potential missed opportunities for optimization.
Innovation Solution
A post-modeling category merging engine that identifies valid pairs of categorical predictors, tests merge strategies to minimize accuracy loss, and merges categories based on empirical evidence to form hybrid categories, thereby refining and optimizing the predictive model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If category merging is performed to simplify the model, then model complexity is reduced, but accuracy may be lost
Solution Approach 1:
The patent applies partial action by selectively merging only certain category pairs that meet specific criteria (low importance scores, high similarity) rather than merging all categories. This partial merging approach reduces model complexity while preserving accuracy-critical distinctions, resolving the contradiction between simplification and accuracy maintenance.
Solution Approach 2:
The patent changes parameters by introducing importance scores and similarity metrics as new evaluation dimensions for category merging decisions. By using these parameter changes to guide selective merging, the system can reduce complexity through merging while using the parameter thresholds to prevent accuracy loss, thus resolving the contradiction.
2Measurement precision
If all possible category pairs are tested for merging, then optimal merger can be found, but computational resources are consumed
Solution Approach 1:
The patent segments the category evaluation process into distinct stages: first calculating importance scores for all categories, then identifying candidate pairs based on similarity, and finally testing only those candidates for merging. This segmentation reduces computational resources by avoiding exhaustive testing of all possible pairs while still finding optimal mergers through the structured multi-stage approach.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating importance scores and similarity metrics for all categories before the merging test phase. This preliminary computation filters out unlikely candidates in advance, so that the actual merging tests focus only on promising pairs, thereby reducing overall computational resource consumption while maintaining optimization accuracy.
3Measurement precision
If categorical variables are analyzed in detail, then feature evaluation improves, but model processing time increases
Solution Approach 1:
The patent extracts and focuses analysis only on categorical variables that have low importance scores and high similarity, rather than analyzing all categorical variables in detail. By taking out only the relevant subsets for detailed examination, the system improves feature evaluation accuracy for critical categories while minimizing processing time by avoiding detailed analysis of already-well-understood categories.
Data Source
AI summary
An embodiment identifies, by a post-modeling category merging engine, a plurality of valid pairs associated with a categorical predictor, the plurality of valid pairs representing potential mergers of categories associated with a categorical predictor of a predictive model. The embodiment tests, by the post-modeling category merging engine, a merge strategy for the plurality of valid pairs to determine a merger that minimizes a loss in accuracy of the predictive model. The embodiment merges, by the post-modeling category merging engine based on the testing, a valid pair in the plurality of valid pairs to form a hybrid category.