A Minimum Risk Wrapper Algorithm for GeneticallySelecting Imprecisely Observed Features, Applied to the Early Diagnosis of Dyslexia

摘要

A wrapper-type evolutionary feature selection algorithm, able to use imprecise data, is proposed. This algorithm is based on a new definition of a minimum Bayesian risk k-NN estimator for vague data. Our information about the risk is assumed to be fuzzy. Therefore, the risk is minimized by means of a modified multicriteria Genetic Algorithm, able to optimize fuzzy valued fitness functions. Our algorithm has been applied to interval-valued data, collected in a study about the early diagnosis of dyslexia. We were able to select a low number of tests that are relevant for the diagnosis, and compared this selection of factors to those sets obtained by other crisp and imprecise data-based feature selection algorithms.