Feature Selection Using Hybrid Evaluation Approaches Based on Genetic Algorithms

For a given set of samples, a new model is proposed to reduce input feature space, which decreases the learning time of classifiers, but also, improves the prediction accuracy according to the chosen relevance criterion. This model is constructed by decision trees and genetic algorithms, which evaluates by means of k nearest neighbor rule for classification, allowing the evolution model parameters of used genetic algorithm. The training set corresponds to the extracted features from pathological (hypernasality) and nonpathological (normal) speech, acquired from 90 children, 45 examples per class. A comparative analysis between different approaches about feature selection is performed upon experimental results, showing the feasibility of this approach in such a cases involving pathologies recognition.