A Genetic Algorithm Based Feature Selection Wrapper

Abstract

Recommender systems are typically hybrid models that employing several differenttechniques including neighborhood based algorithms. One of the maindesign issues in such neighborhood models is the determiningthe relative importance of different features when computingsimilarities. The purpose of this thesis explores the use of a geneticalgorithm based wrapper to determine optimal weights in terms oftheir predictive accuracy. The method shows improved performance compared to unweighted models using the same feature set.