Variable Influence Analysis Across Periodic Regression Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In semiconductor and chemical plants, existing methods for analyzing quality characteristic variations using regression models struggle to accurately identify influential factors due to noise in small datasets and overlook factors with temporary or steady influences, leading to incomplete ranking and factor identification.
Innovation Solution
An information processing device calculates the degree of influence and selection frequency of explanatory variables over multiple periods, categorizing them into four categories to facilitate accurate identification of factors influencing quality characteristics, including those with high or low influence and frequency, and displays these in a matrix diagram for user extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If regression analysis is performed using small datasets, then model estimation can be completed quickly, but the analysis accuracy deteriorates due to noise in the data
Solution Approach 1:
The patent divides the analysis period into multiple segments (first period and second period) and performs separate regression analyses for each segment. This segmentation allows the system to process smaller datasets individually, reducing noise impact while maintaining computational efficiency. The results from multiple segments are then integrated to provide comprehensive factor identification.
2Loss of time
If only the latest data is used for model estimation, then the analysis can be performed quickly with current information, but factors with steady influences over time are overlooked
Solution Approach 1:
The patent performs regression analysis on historical data (first period) in advance before analyzing the latest data (second period). This preliminary action preserves information about steady influences that may not be apparent in recent data alone. The pre-computed results from historical analysis are then combined with latest data analysis to provide comprehensive factor identification.
3Measurement precision
If multiple regression models are estimated over multiple periods, then comprehensive factor identification is achieved, but the computational complexity and processing time increase
Solution Approach 1:
The patent segments the overall analysis into distinct time periods with separate regression models. By dividing the comprehensive analysis into manageable periodic segments, the system achieves thorough factor identification while keeping each individual model estimation computationally tractable. The segmented approach balances accuracy with processing efficiency.
4Measurement precision
If all explanatory variables are analyzed simultaneously, then complete factor identification is attempted, but the complexity of analysis increases and important factors are difficult to identify
Solution Approach 1:
The patent applies segmentation by analyzing explanatory variables across different time periods separately rather than all at once. Each periodic regression model focuses on a specific time segment, reducing the complexity of simultaneous analysis. This segmented approach makes it easier to identify important factors in each period while maintaining comprehensive coverage through aggregation of results.
Data Source
AI summary
According to an embodiment, an information processing device includes one or more processors. The processors calculate a first degree of influence of a plurality of variables on output data, and a frequency at which the plurality of variables are selected as a variable influencing the output data, based on K first models. The K first models are models estimated using a plurality of pieces of input data including the plurality of variables. The plurality of input data are obtained in K periods. K is an integer of 2 or more. The first model receives input of the input data including the plurality of variables and outputs the output data. The processors output the first degree of influence and the frequency in association with each other.