A novel benchmark dataset for functional data analysis and an initial comparative analysis

F. Heinrichs, T. Vasconcelos Afonso

Most methods for functional data require the data to be registered, that is, aligned in time. While this assumption simplifies the analysis, it cannot be justified in many applications, especially when working with continuously recorded sensor data, where no "start", "end" or any other landmark exists.

We present a novel dataset that can be used for benchmark experiments and comparisons of different FDA methods. The dataset contains eye-tracking and EEG data of more than 100 participants, measured via camera and simultaneously with an EEG headset. Generally, the task is to reconstruct eye movements from the EEG. The dataset contains different levels of difficulty (registered and unregistered data, continuous and abrupt movements, four directions and unrestricted movement).

Additionally, we present a comparative analysis of functional neural networks, that are specifically designed for unregistered data, with established methods.

Keywords: Benchmark dataset, EEG data, eye-tracking

Functional data analysis II
June 12, 2025  11:30 AM
Sala VIP Jaume Morera i Galícia

M. Vidal, A. M. Aguilera

Y. Larriba, I. Fernández, C. Canedo Ortega, C. Rueda

B. Pulido Bravo, A. M. Franco-Pereira, R. E. Lillo

M. C. Aguilera-Morillo, M. Durban, P. Hernández-Amaro