Data Analysis Method for Imbalanced Datasets Using Weighted Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data analysis methods, such as regression equations, face accuracy issues when dealing with imbalanced datasets, where positive cases outnumber negative cases, leading to poor prediction performance and improper evaluation of known data.
Innovation Solution
A data analysis process that randomly extracts equivalent numbers of positive and negative case measured values, generates a prediction equation, calculates predicted values for remaining cases, and computes a weighted correct answer percentage to assess prediction accuracy, allowing for the adjustment of weight variables based on the imbalance ratio.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all data are used to generate the prediction equation, then the evaluation of known data appears correct, but the prediction accuracy deteriorates when positive cases overwhelmingly outnumber negative cases
Solution Approach 1:
The patent divides the dataset into two separate sets: a first set containing only positive case data and a second set containing only negative case data. This segmentation allows independent analysis of each class, preventing the imbalance between positive and negative cases from degrading prediction accuracy. The prediction equation is then generated by combining results from both segmented analyses.
2Adaptability or versatility
If conventional regression equations are used, then data analysis can be performed, but the prediction accuracy drops when values for positive cases are overwhelmingly greater than values for negative cases
Solution Approach 1:
The patent applies different analysis methods to different parts of the data. Specifically, it uses a first analysis method for positive case data and a second analysis method for negative case data, allowing each method to be optimized for its specific class characteristics. This local quality approach enables the system to handle imbalanced data effectively while maintaining high prediction accuracy for both positive and negative cases.
3Ease of operation
If all data are used for evaluation, then the evaluation process is simple, but the predicted value is improperly evaluated to be correct in an excessive manner
Solution Approach 1:
The patent extracts and separates the evaluation process from the training process by using distinct data sets. The first set of positive case data and second set of negative case data are used for generating the prediction equation, while a separate evaluation set is used for assessing performance. This extraction prevents overevaluation of known data and provides a reliable, unbiased assessment of prediction accuracy.
Data Source
AI summary
A non-transitory computer-readable medium stores a program that causes a computer, which has a memory device storing a set of measured values that included a set of positive case measured values and a set of negative case measured values, to execute a process. The process includes extracting randomly, a positive case measured value group and a negative case measured value group from the set of measured values; generating based on the positive case measured value group and the negative case measured value group, a prediction equation that predicts the objective variable for a prediction algorithm; first calculating a first predicted value group; second calculating a second predicted value group; first identifying a first coincident-case count of predicted values; second identifying a second coincident-case count of predicted values; computing a weighted correct answer percentage; and outputting a computation result obtained at the computing.