SOME NEW METHODS FOR SUPERVISED CLASSIFICATION FOR FUNCTIONAL DATA

Abstract

Functional data are getting prevalent in many research and industrial fields in recent decades. It is often of interest to classify functional data properly. A number of supervised classification methods have been proposed in the literature. In this thesis, I propose and study three new classifiers for supervised classification for functional data, including a supervised classification method based on functional cosine similarity, a new k-nearest neighbor classifier, and an inverse distance based classifier. Intensive simulation studies and a number of real functional data examples are conducted to demonstrate and illustrate the good performance of the three new functional classifiers via comparing them against several existing competitors.