Regression Coefficient Restoration for Multicollinear Performance Indicators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for business processing struggle to accurately determine the impact of performance indicators on business indicators due to the multicollinearity problem, which leads to the inability to calculate the regression coefficient of removed performance indicators.
Innovation Solution
A method that involves obtaining multiple performance indicators associated with a business indicator, identifying a target performance indicator with a multicollinearity problem, determining a first performance indicator with a high correlation to the target indicator, calculating the regression coefficient between the target and first performance indicators, and restoring the regression coefficient of the target performance indicator to the business indicator using the first regression coefficient and other performance indicators' coefficients.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If linear regression processing is performed on performance indicators with high correlation, then multicollinearity problem occurs, but if one indicator is removed to avoid multicollinearity, then the regression coefficient of the removed indicator cannot be determined
Solution Approach 1:
The patent performs preliminary action by calculating the regression coefficient between the target performance indicator and first performance indicator before removing the target indicator for multicollinearity processing. This pre-calculated coefficient is then used to restore the target indicator's regression coefficient later, ensuring accurate determination without multicollinearity issues
Solution Approach 2:
The patent applies discarding and recovering by temporarily removing the target performance indicator from the regression model to avoid multicollinearity, then recovering its regression coefficient through calculation using the pre-stored coefficient and other regression coefficients. This allows the indicator to be excluded during processing but its impact to be recovered and determined accurately
2Measurement precision
If performance indicators with high correlation are retained in the regression model, then multicollinearity problem occurs affecting accuracy, but if removed, then loss of information about their impact on business indicators
Solution Approach 1:
The patent extracts the regression coefficient relationship between the target performance indicator and first performance indicator from the multicollinearity-affected model. By taking out this specific coefficient relationship, the patent can determine the target indicator's impact on business indicators without the distorting effect of multicollinearity
Solution Approach 2:
The patent uses the first performance indicator as an intermediary to determine the target performance indicator's regression coefficient. The first indicator serves as a mediator that has high correlation with the target indicator, allowing the target indicator's impact to be determined through the intermediary without direct inclusion in the regression model
Data Source
AI summary
Embodiments of the disclosure provide a method, apparatus and electronic device for business processing. The method includes: obtaining a plurality of performance indicators associated with a business indicator; determining, in the plurality of performance indicators, a target performance indicator with a multicollinearity problem; determining a first performance indicator; determining a first regression coefficient between the target performance indicator and the first performance indicator; determining a plurality of second regression coefficients between the business indicator and a plurality of second performance indicators; and restoring a regression coefficient of the target performance indicator to the business indicator based on the first regression coefficient and the plurality of second regression coefficients, the regression coefficient of the target performance indicator to the business indicator being directly proportional to a degree of impact of the target performance indicator on the business indicator.