Indicator Selection Using Machine Learning Influence Degree
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for evaluating targets using multiple indicators are inefficient in selecting the appropriate combination of indicators for accurate evaluation.
Innovation Solution
An indicator selection apparatus and method that acquires category variable specification information and selects support variables by generating a model using machine learning, determining the influence degree of combination variables on the model's accuracy, and choosing indicators based on this degree.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple indicators are combined to evaluate an evaluation target, then the accuracy of evaluation is improved, but the difficulty of selecting appropriate indicator combinations increases
Solution Approach 1:
The patent replaces manual indicator selection with an automated machine learning system. The selection unit automatically generates combination variables by combining category variables with other indicators, trains multiple models with different combinations, and selects the optimal combination based on model performance metrics, eliminating the need for manual trial-and-error in indicator selection.
Solution Approach 2:
The patent systematically varies the parameters of indicator combinations by generating multiple combination variables with different pairings of category variables and other indicators. The machine learning models are trained with these varying parameter combinations, and the optimal parameters are identified based on model accuracy metrics.
2Adaptability or versatility
If the number of feature values is increased by combining multiple explanatory variables, then the evaluation capability is improved, but the number of feature values becomes too great to manage
Solution Approach 1:
The patent segments the feature generation process by first identifying a specific category variable, then systematically combining it with other indicators to generate a controlled set of combination variables. This segmentation approach prevents the explosive growth of feature values by organizing the combination process around specific category variables rather than generating all possible combinations.
Solution Approach 2:
The patent generates combination variables by partially combining indicators - specifically pairing a category variable with other relevant indicators rather than combining all possible indicator pairs. This partial action approach generates sufficient feature values for accurate evaluation without creating an unmanageable quantity.
3Adaptability or versatility
If manual selection of indicator combinations is performed, then flexibility in choosing indicators is maintained, but the time and effort required for selection increases
Solution Approach 1:
The patent implements self-service by enabling the system to automatically perform indicator selection without human intervention. The selection unit autonomously generates combination variables, trains multiple machine learning models, evaluates their performance, and selects the optimal indicator combination based on predetermined criteria, making the system self-sufficient in the selection process.
Solution Approach 2:
The patent incorporates feedback mechanisms where the performance of each machine learning model is evaluated based on metrics such as accuracy, precision, and recall. This feedback is used to iteratively refine the selection process and identify the optimal indicator combination, allowing the system to learn from and improve upon previous selections.
Data Source
AI summary
An indicator selection apparatus (10) includes an acquisition unit (110) and a selection unit (120). The acquisition unit (110) acquires information (hereinafter, described as category variable specification information) that specifies a category variable. The selection unit (120) selects a support variable from among the plurality of indicators described above for each category variable. Specifically, the selection unit (120) generates a first model by performing machine learning by using a combination (combination variable) of the category variable and the support variable as an explanatory variable and using an evaluation result of an evaluation target as an objective variable. Then, a first influence degree is generated for each of a plurality of the combination variables. The first influence degree indicates magnitude of an influence of the combination variable on accuracy of the first model. Then, the selection unit (120) selects a support variable by using the first influence degree.