Network Sensitivity Analysis for Machine Learning Parameter Importance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models, especially non-linear models like neural networks, lack the ability to effectively determine the relative contribution of individual input parameters to their outputs, making it difficult to understand and visualize the impact of each parameter on the model's predictions.
Innovation Solution
The method involves re-ordering the columns of the input matrix based on the values in each row, separating them into sub-matrices, and then inputting these sub-matrices into a fully-trained machine learning model to generate outputs, which are subsequently normalized to produce a network sensitivity analysis (NSA) curve, allowing for the visualization of parameter contributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-linear machine learning models are used to improve prediction accuracy, then the model's ability to detect hidden trends and process large data sets is enhanced, but the ability to determine the relative contribution of individual input parameters is lost
Solution Approach 1:
The patent segments the input data into multiple groups based on parameter values and processes each group separately through the machine learning model. This segmentation allows the system to maintain the non-linear prediction capabilities while also analyzing the contribution of individual parameters by comparing outputs across different parameter value groups.
Solution Approach 2:
The patent introduces an intermediary analysis layer that sits between the non-linear machine learning model and the final output interpretation. This intermediary layer processes the model outputs to extract parameter contribution information, enabling the system to retain both prediction accuracy and interpretability without compromising either aspect.
2Measurement precision
If complex non-linear models are applied to analyze large data sets, then hidden trends can be detected, but the complexity of determining parameter contributions increases
Solution Approach 1:
The patent divides the complex analysis task into segments by grouping input data according to parameter values and processing each segment separately. This segmentation simplifies the overall complexity by breaking down the parameter contribution analysis into manageable portions that can be handled systematically.
Solution Approach 2:
The patent performs preliminary actions by pre-grouping the input data into distinct groups based on parameter values before feeding them into the machine learning model. This preliminary organization simplifies the subsequent analysis by preparing the data in a structured manner that facilitates easier parameter contribution determination.
3Ease of operation
If the input data is processed in its original form, then the analysis is straightforward, but the ability to visualize and interpret parameter impacts is limited
Solution Approach 1:
The patent segments the input data into multiple groups based on parameter values and processes each group separately. This segmentation transforms the raw data into an organized structure that maintains processing simplicity while enabling clear visualization and interpretation of parameter impacts through systematic comparison of grouped results.
Solution Approach 2:
The patent adds another dimension to the data structure by introducing group identifiers and organized categorizations alongside the original data features. This dimensional enhancement preserves the simplicity of processing the original data while adding interpretability dimensions that enable visualization of parameter impacts through structured output organization.
Data Source
AI summary
A computer-implemented method to establish a relative importance of an input parameter pj in a plurality of input parameters pi in a data set input to a machine learning model, the data set represented by a j row by k column matrix Im, an intersection of each row with each column defining an element, the method includes for each of the plurality of parameters pi in the input data set, a computer sorts columns ki of the matrix Im. to produce a re-ordered matrix Im,j; the computer determines a hyper-parameter N* of sub-matrices into which may be sorted the values in a jth row of the re-ordered matrix Im,j; the computer generates a plurality of group sub-matrices Gi, each of the group sub-matrices comprising a subset of columns and the jth row; the computer inputs the re-ordered matrix Im,j into a fully-trained machine learning model to produce machine learning model outputs; and the computer produces normalized mean values of the outputs.