Home // HEALTHINFO 2024, The Ninth International Conference on Informatics and Assistive Technologies for Health-Care, Medical Support and Wellbeing // View article

Does Complexity Pay Off? Applying Advanced Algorithms to Depression Detection on the GLOBEM Dataset

Authors:
Sebastian Cavada
Alvaro Berobide
Yevheniia Kryklyvets

Keywords: Depression Detection; Time-Series Analysis; Deep Learning; Domain Generalization; Mental Health.

Abstract:
This manuscript evaluates the performance of state- of-the-art time series analysis algorithms for depression detection on the Generalization of LOngitudinal BEhavior Modeling (GLOBEM) dataset. We assess Time-Series Mixer (TSMixer), Cross- former, Gated Recurrent Unit (GRU), Convolutional Neural Network with Long Short-Term Memory (CNN_LSTM) and introduce a novel self-developed algorithm with the goal of increasing accuracy over the original Reorder. While these models demonstrate robust out-of-domain generalization, they fail to surpass the accuracy of the baseline Reorder algorithm, which was specifically developed for in-domain analysis by the GLOBEM team. Our findings reveal consistently low performance across all models, suggesting limitations inherent in the dataset rather than the algorithms themselves. We hypothesize that the dataset’s absence of critical variables and insufficient granularity likely limits model convergence. This hypothesis is supported by similar studies that achieved higher accuracy using more frequent data points with similar architecture approaches. Based on these insights, we suggest that future studies might benefit from incorporating more granular sensor measurements and more sophisticated data types, such as, but not limited to, Heart Rate Variability (HRV).

Pages: 48 to 52

Copyright: Copyright (c) IARIA, 2024

Publication date: November 3, 2024

Published in: conference

ISSN: 2519-8491

ISBN: 978-1-68558-204-3

Location: Nice, France

Dates: from November 3, 2024 to November 7, 2024