Knowledge Discovery in Heterogeneous Federated Learning: Encoded Multimodal Features for Depression Prediction
Keywords:
Knowledge discovery, Federated learning, encoded feature, artificial intelligence, depressionAbstract
Depression is a prevalent and serious condition; however, developing reliable knowledge discovery models for early detection remains a significant challenge. Two major barriers hinder progress: (1) stringent privacy regulations that restrict sharing of patient data across institutions, and (2) substantial variability in how hospitals record clinical information—both in content and format. As a result, most research relies on single-institution datasets, which limits the generalizability of the discovered knowledge to other clinical contexts.
We propose a privacy-preserving federated learning for multi-hospital collaboration that avoids raw data sharing. By converting institutional records into standardized 'feature codes'—compact numerical representations of clinical patterns—our method ensures data utility while safeguarding patient privacy.
We evaluated three heterogeneous depression datasets: workplace-related, sleep/nutrition-related, and socioeconomic-related. Our federated learning model demonstrated superior performance across multiple metrics - achieving 94% accuracy (vs. institutional average 91%), 0.95 AUC (vs. 0.84), and 91 precision (vs. 74) - confirming its viability for privacy-preserving cross-institutional knowledge discovery in healthcare.
Downloads
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 International Congress of Knowledge and Innovation
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
DECLARATION OF ASSIGNMENT AND TRANSFER OF PATRIMONIAL RIGHTS ON ARTICLE PUBLISHED IN THE CIKI PROCEEDINGS AND AUTHORIZATION FOR PUBLICATION
The AUTHOR, according to the law n. 9.610 of February 19th, 1998, hereby declares to whomsoever may concern, that it assigns and transfers, in a universal, definitive, irreversible, exclusive and gratuitous manner, all of its author's economic rights on the article submitted to the International Congress of Knowledge and Innovation - ciki.
The AUTHOR guarantees:
- That the article is original, except for the citations of other published works, provided that the limitations expressed in articles 46 and 47 of Law 9.610 of February 19, 1998 are observed;
- That the article does not contain any slanderous or defamatory statements and does not infringe any intellectual, commercial or industrial property rights of any third parties;
- To promptly compensate the International Congress of Knowledge and Innovation - ciKi for any indemnities, losses or expenses arising from the breach of the guarantees expressed in paragraphs 1 and 2, above.
With this assignment and transfer of the patrimonial rights referring to the author's right, the International Congress of Knowledge and Innovation - ciKi and its successors are free of any copyright payment to the AUTHOR or to their heirs or successors.
The AUTHOR further declares that the International Congress of Knowledge and Innovation - ciKi is fully authorized to use this article, in whole or in part, edited or complete, in Portuguese and in all other languages, in print, in Electronic means, internet, for commercial purposes or not, including distributing, adapting, creating derivative works, assigning rights to third parties in Brazil and / or abroad, including but not limited to: teaching, study and research; publication and dissemination; quote; use in general telecommunication means; audiovisual use in general, including all existing or future digital technologies, capable of storing and reproducing data.
The AUTHOR has guaranteed the moral rights over his article, including the binding of his name as author of the article object of this transfer.
The AUTHOR must always make a written request to the International Congress of Knowledge and Innovation - ciKi, when he intends to use his article, and is obliged to always insert the credit in the original publication, citing the bibliographic reference, complete and legible.
