Model-free Clustering for Machine Learning Interpretability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, especially those with hidden layers like Neural Networks and Support Vector Machines, face challenges in interpretability, making it difficult to understand predictions and aligning models with problem solutions, as existing methods are not effective in extracting insights from data with high prediction confidence.
Innovation Solution
A method for model-free clustering of data that identifies high prediction confidence data groups, transforms them using Local Interpretable Model-agnostic Explanations (LIME) to select high rank predictors, and clusters these transformed data groups to generate insights about the underlying model, improving interpretability and avoiding the impact of non-relevant predictors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If model-free clustering is applied to data with high prediction confidence, then interpretability of machine learning models is improved, but the complexity of the clustering process increases
Solution Approach 1:
The patent segments the data into high prediction confidence groups before clustering, and further segments the feature space by identifying and selecting only high-rank predictors for each data case. This segmentation reduces the complexity of analyzing the entire dataset by focusing only on relevant portions and features, thereby improving interpretability without proportionally increasing overall process complexity.
Solution Approach 2:
The patent extracts only the high-rank predictors from the full feature set for each data case, and extracts only high prediction confidence data points from the full dataset. This extraction eliminates redundant and less relevant information, simplifying the clustering process while enhancing the interpretability of the resulting models.
2Measurement precision
If all predictors are used in clustering, then comprehensive data analysis is achieved, but non-relevant predictors reduce clustering accuracy
Solution Approach 1:
The patent applies local quality by identifying and selecting high-rank predictors specifically for each data case based on their local importance to that case's prediction. This means different sets of predictors are selected for different data cases, optimizing clustering accuracy for each local region rather than using a fixed set of all predictors, thereby improving precision without uniformly increasing the quantity of predictors.
Solution Approach 2:
The patent uses partial action by selecting only the necessary high-rank predictors for each data case rather than using all available predictors. This partial approach focuses computational resources on the most relevant features, improving clustering accuracy while avoiding the excessive action of processing all predictors, which would reduce accuracy due to noise and irrelevance.
Data Source
AI summary
A method of clustering data generated by an unknown model, the method including accessing the data, wherein the data includes a prediction target and a confidence, extracting a data group with high prediction confidence from the data, wherein the data group comprises a plurality of data cases, and where each of the data cases is described by a plurality of predictors, identifying high rank predictors of each the data cases in the data group, transforming the data group into a transformed data group including only the high rank predictors for each of the data cases, wherein the high rank predictors are ranked within each of the data cases included in the transformed data group, clustering the transformed data group to generate clusters, and profiling the clusters to extract an insight about the unknown model.