Supervised Learning via Ensembles of Diverse Functional Representations: the Functional Voting Classifier

Many conventional statistical and machine learning methods face challengeswhen applied directly to high dimensional temporal observations. In recentdecades, Functional Data Analysis (FDA) has gained widespread popularity as aframework for modeling and analyzing data that are, by their nature, functionsin the domain of time. Although supervised classification has been extensivelyexplored in recent decades within the FDA literature, ensemble learning offunctional classifiers has only recently emerged as a topic of significantinterest. Thus, the latter subject presents unexplored facets and challengesfrom various statistical perspectives. The focal point of this paper lies inthe realm of ensemble learning for functional data and aims to show howdifferent functional data representations can be used to train ensemble membersand how base model predictions can be combined through majority voting. Theso-called Functional Voting Classifier (FVC) is proposed to demonstrate howdifferent functional representations leading to augmented diversity canincrease predictive accuracy. Many real-world datasets from several domains areused to display that the FVC can significantly enhance performance compared toindividual models. The framework presented provides a foundation for votingensembles with functional data and can stimulate a highly encouraging line ofresearch in the FDA context.