Feature selection algorithms for very high dimensional data and mixed data

Abstract

Feature selection is an important issue in pattern recognition. The goal of feature selection algorithm is to identify a set of relevant features, based on which to construct a classifier for a pattern recognition problem. This thesis addresses the problem of feature selection for very high dimensional data and mixed data, which exist in many application domains of pattern recognition nowadays. The proposed feature selection algorithms aim to eliminate both irrelevant and redundant features while retaining major discriminating underlying data.