Regression Model Similar Feature Extraction for Collinearity

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Penalized regression models fail to extract features similar to selected explanatory variables, leading to potential overlooking of important factors in regression analysis, especially when dealing with numerous explanatory variables in industrial processes like water supply and sewerage, power, and chemical plants.

Innovation Solution

An analysis apparatus that includes a regression model constructor and a similar feature extractor, which constructs a regression model and calculates the similarity degree between features to extract similar features with a predetermined threshold, aiding in identifying influential factors that may be overlooked due to collinearity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If penalized regression model is used to automatically extract features from numerous explanatory variables, then feature selection efficiency is improved, but important factors are likely to be overlooked due to inability to extract similar features

Engineering Contradiction:

Improvefeature selection efficiencyVSAvoidcompleteness of important factors

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent creates a 'similar feature' as a copy or representation of the selected explanatory variable. By calculating similarity degrees between selected variables and unselected variables, the system identifies unselected variables that are highly similar to selected ones, effectively creating a copied representation of important features that the penalized regression model may have excluded.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent adds a new dimension of analysis by introducing the concept of 'similarity degree' alongside the traditional regression coefficients. This additional dimension allows the system to evaluate features not only based on their direct regression weights but also based on their similarity to selected features, thereby capturing important factors that might be overlooked in the original penalized regression analysis.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Productivity

If penalized regression model selects explanatory variables, then model construction speed is improved, but collinearity causes important factors to be overlooked

Engineering Contradiction:

Improvemodel construction speedVSAvoiddetection of influential factors

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system implements a feedback mechanism where the results of penalized regression are used to identify selected variables, which then serve as the basis for calculating similarity degrees. This feedback loop allows the system to iteratively refine feature selection by considering both the regression model's direct selections and the similarity-based expansions, ensuring that collinear but important factors are not overlooked.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent performs preliminary penalized regression analysis to select initial explanatory variables before conducting the similarity degree calculation. This preliminary action establishes a foundation of selected variables that guide the subsequent identification of similar features, allowing the system to efficiently narrow down the search space while maintaining comprehensive coverage of important factors.

Inventive Principle:

Principle #10Preliminary action

3Ease of operation

If only selected explanatory variables are analyzed, then analysis simplicity is improved, but false negatives occur in factor identification

Engineering Contradiction:

Improveanalysis simplicityVSAvoidaccuracy of factor identification

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The patent segments the feature analysis into two distinct parts: selected explanatory variables and similar features. This segmentation allows the system to maintain simplicity by clearly defining two categories of features to analyze, while simultaneously improving reliability by ensuring that both direct selections and similar but unselected features are considered in the final analysis.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system merges the results from penalized regression selection with the similarity-based feature identification. By combining these two approaches, the patent creates a comprehensive feature set that includes both directly selected variables and their similar counterparts, thereby eliminating false negatives while maintaining analytical clarity through the structured two-category framework.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11216741B2Analysis apparatus, analysis method, and non-transitory computer readable medium

Publication Date: 2022.01.04 KK TOSHIBA

AI summary

A regression analysis apparatus includes a regression model constructor and a similar feature extractor. A regression model constructor is configured to construct a regression model that represents an objective variable with a plurality of explanatory variables that correspond to any one of a plurality of features and with a regression coefficient of the plurality of explanatory variables by performing regression analysis using analysis target data including the plurality of features with one of the plurality of features as the objective variable. A similar feature extractor is configured to calculate a similarity degree between a feature other than a feature that corresponds to the objective variable in the analysis target data and the plurality of explanatory variables, and each of the plurality of explanatory variables, and configured to extract a similar feature having the similarity degree higher than a predetermined value.