Memetic algorithms for feature/gene selection

Abstract

This dissertation presents novel memetic frameworks for the hybridization of wrapper and filter feature selection methods on classification problems. The frameworks incorporate filter methods in the traditional genetic algorithm (GA) to improve classification performance and accelerate the search in identifying the core feature subsets. Particularly, the filter methods are introduced to add or delete features from a candidate feature subset encoded in a GA solution. Using memetic frameworks, we propose and systematicall study three feature selection algorithms, Wrapper-Filter Feature Selection Algorithm (WFFSA), Markov Blanket Embedded Genetic Algorithm (MBEGA), and Markov Blanket Embedded Multiobjective Memetic Algorithm (MBE-MOMA).