Genetic Algorithm Feature Selection for EEG Dementia Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting beta amyloid in dementia patients, such as amyloid PET tests, are expensive and not guaranteed to provide a positive result for medication prescription, while electroencephalogram (EEG) is a low-cost but less effective method for diagnosing dementia due to the need to identify optimal EEG channels and frequencies for disease prediction.
Innovation Solution
A machine learning method using a genetic algorithm for feature selection that constructs a high-accuracy learning model by defining feature sets including EEG occurrence locations and frequency bands, generating feature combinations, calculating prediction accuracy, and updating the feature set to determine good features for disease classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If EEG is used for disease diagnosis, then cost is reduced, but diagnostic accuracy deteriorates due to difficulty in identifying optimal channels and frequencies
Solution Approach 1:
The patent applies preliminary action by pre-defining a structured feature set that includes specific EEG channels (e.g., Fp1, Fp2, F3, F4, C3, C4, P3, P4, O1, O2) and frequency bands (delta, theta, alpha, beta, gamma) before the actual diagnosis process. This preliminary organization of features into a systematic framework enables the genetic algorithm to efficiently evaluate and select the most discriminative features, thereby improving diagnostic accuracy while managing complexity.
Solution Approach 2:
The patent replaces the manual or heuristic feature selection process with a genetic algorithm, which is an automated computational method inspired by natural selection. This substitution allows the system to automatically identify optimal EEG channels and frequency bands without manual intervention, significantly improving diagnostic accuracy while the automation manages the complexity of feature selection.
2Measurement precision
If all EEG channels and frequencies are analyzed, then diagnostic accuracy improves, but computational complexity increases
Solution Approach 1:
The patent extracts only the most relevant and discriminative features from the complete set of EEG channels and frequency bands using a genetic algorithm. Instead of analyzing all possible features, the algorithm evaluates each feature's contribution to disease classification and selects only those that provide the most information, thereby achieving high measurement precision while reducing the effective search space and computational complexity.
Solution Approach 2:
The patent applies partial action by selecting a subset of the most important EEG features rather than analyzing all channels and frequencies. The genetic algorithm determines the optimal number and combination of features needed for accurate diagnosis, performing only the necessary computations to achieve the required diagnostic precision without the excessive computational burden of analyzing every possible feature.
3Reliability
If feature dimension increases, then model accuracy improves, but training time increases
Solution Approach 1:
The patent performs preliminary feature selection using a genetic algorithm before the actual classification training. By pre-identifying and selecting the most discriminative features in advance, the system reduces the dimensionality of the data that needs to be processed during model training. This preliminary action maintains high classification accuracy while significantly reducing the training time required for the machine learning model.
Solution Approach 2:
The patent uses partial action by selecting only the essential features needed for accurate disease classification rather than using all available features. The genetic algorithm determines the optimal feature subset that provides sufficient information for high accuracy, avoiding the excessive training time that would result from processing all possible features while maintaining reliable classification performance.
Data Source
AI summary
The present disclosure relates to a machine learning method and apparatus using steps feature selection based on a genetic algorithm, and the machine learning method includes defining a feature set including a plurality of features, generating a plurality of feature combinations including n-dimensional features (n is a natural number) for the feature set, independently constructing feature models for the plurality of feature combinations and calculating prediction accuracy for each of the feature models as a prediction result for a predetermined data set, arranging the feature models according to the prediction accuracy to determine at least one good feature model that satisfies a preset criterion, determining at least one good feature from among features included in a corresponding feature set of the at least one good feature model, and updating the feature set to include only the at least one good feature and re-determining a good feature model for a (n+1)-dimensional feature combination based on the updated feature set.