Decision Trees with Smoothed Data

This study aims to compare raw and functional data within a decision tree framework. The study examines the patterns exhibited by countries across various variables based on the number of deaths per million during the COVID-19 pandemic. Vectorial inputs are transformed into functional inputs, and a smoothed dataset representing the raw data with fewer functional observations is obtained. Regression and classification trees obtained from smoothed and unsmoothed data are then compared. Additionally, changes in the decision tree model are examined with the functions obtained by changing the number of basis functions. Using the functional data approach, the importance of variables affecting mortality rates is identified, and changes in prediction results arising from the transformation of the data structure from raw to functional form are analyzed. The main contribution of the study is the implementation of decision tree analyses using functional objects. The findings indicate that fitting the data through a basis function approach provides higher classification performance than using raw data. Moreover, functional data–based trees achieve comparable or higher accuracy with simpler and more interpretable tree structures. Based on these findings, classification decision tree analysis using smoothed data is proposed as an alternative approach for the analysis of functional predictors.