Leveraging Functional Diversity in Statistical Ensemble Learning for Robust Functional Data Classification

This paper introduces the Randomized Spline Trees (RST) algorithm for clas- sifying environmental time series. RST extends ensemble methods to functional data by introducing randomization in B-spline parameters, thus inducing functional diversity. Experimental results on six UCR environmental time series datasets show that RST variants often outperform standard Random Forests and Gradient Boosting. The findings confirm the effectiveness of diversity in functional data ensembles, as also highlighted by prior re- search on the Functional Voting Classifier (FVC).