Genetic Algorithm Feature Selection Using Classifier Ensemble Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Genetic algorithm-based feature selection in medical datasets is hindered by noise and bias due to random data splitting, leading to inaccurate evaluation of feature subsets and potential discarding of optimal subsets, especially in datasets with limited samples and large feature sets.
Innovation Solution
Integrating classifier ensemble methods into the genetic algorithm process by applying multiple data splitting patterns to build multiple classifiers, combining their results to form an integrated accuracy value used as a fitness value for feature subsets, and using re-sampling techniques to generate diverse training sets, thereby reducing noise impact and improving feature selection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If random data splitting is used in genetic algorithm-based feature selection, then the evaluation process is simple and fast, but noise and bias are introduced leading to inaccurate evaluation of feature subsets
Solution Approach 1:
The patent divides the evaluation process into multiple independent evaluations using different data splitting patterns (e.g., random split, stratified split, k-fold cross-validation). Each splitting pattern provides an independent assessment of the feature subset, and the results are aggregated to produce a more reliable overall evaluation metric that reduces noise and bias from any single split.
Solution Approach 2:
The patent combines multiple classification results from different data splitting patterns into an integrated accuracy value. By merging the evaluations from multiple classifiers trained on different data partitions, the system achieves a more robust and accurate assessment of feature subset performance that is less sensitive to random variations in any single split.
2Measurement precision
If multiple data splitting patterns and multiple classifiers are used to build an integrated accuracy result, then noise-induced bias is reduced and feature selection accuracy is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary actions by pre-defining multiple data splitting patterns and selecting multiple classifier types before the main feature selection process. This allows the system to efficiently evaluate different feature subsets using a prepared framework of evaluation strategies, reducing the computational burden during the actual genetic algorithm optimization by avoiding the need to design evaluation protocols on-the-fly.
3Loss of information
If all available features are used in the dataset, then complete information is available for analysis, but computational infeasibility occurs and overfitting problems arise when data samples are limited
Solution Approach 1:
The patent applies partial action by selectively evaluating only the most relevant feature subsets through the genetic algorithm's fitness function, which guides the search toward promising regions of the feature space. Rather than exhaustively evaluating all possible feature combinations (which would be computationally infeasible), the algorithm efficiently explores a representative sample of potential subsets, achieving good enough solutions with reasonable computational cost.
Data Source
AI summary
Methods for performing genetic algorithm-based feature selection are provided herein. In certain embodiments, the methods include steps of applying multiple data splitting patterns to a learning data set to build multiple classifiers to obtain at least one classification result; integrating the at least one classification result from the multiple classifiers to obtain an integrated accuracy result; and outputting the integrated accuracy result to a genetic algorithm as a fitness value for a candidate feature subset, in which genetic algorithm-based feature selection is performed.