Concurrent Classification of Pseudomonas aeruginosa Occurrence From Routine Municipal Swimming‐Pool Monitoring Data: An Explainable Machine‐Learning Proof‐of‐Concept With Grouped Validation
Water Environment Research · 2026-11 · 98 (11)
DOI: 10.1002/wer.70653 ↗Kurz gefasst
Fachartikel in der Zeitschrift Water Environment Research, online erschienen am 11.11.2026. Verfasst von Athanasia Palaiologou und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Public Health, School of Medicine, University of Patras, Patras, Greece. Hauptthemen laut MEDLINE: Swimming Pools, Pseudomonas aeruginosa, Machine Learning, Environmental Monitoring. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Routine swimming‐pool surveillance may support detection‐oriented screening, but rare positives and repeated sampling complicate model validation. This retrospective study analyzed 453 samples (37 occurrence‐positive) from four chlorinated pools in two Greek facilities (2022–2025). Random forest, logistic regression, gradient boosting, and extra trees were evaluated using concurrent monitoring variables, physicochemical predictors and previous‐sample information. Row‐level, leave‐one‐pool‐out, leave‐one‐facility‐out and temporal validation were compared with a residual‐chlorine‐only baseline. Full‐monitoring random forest achieved PR‐AUC 0.532 within the network, decreasing to 0.386 for held‐out pools and 0.157 for a held‐out facility; corresponding logistic‐regression values were 0.525 and 0.326. Residual chlorine alone achieved PR‐AUC 0.452, with no model significantly outperforming it. Previous‐sample predictors provided no convincing early‐warning signal. Screening thresholds yielded substantial false alerts. These results support exploratory concurrent classification, not prospective warning or deployment readiness, and highlight the need for independent multifacility validation. Machine‐learning models classified Pseudomonas aeruginosa occurrence from routine swimming‐pool monitoring data within the observed network, but performance fell in pools and facilities not used for training, and no model clearly outperformed residual chlorine alone. Previous‐sample data gave no convincing early warning, so prospective multifacility validation is needed.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- Department of Public Health, School of Medicine, University of Patras, Patras, Greece.
ORCID-Profile: Athanasia Palaiologou, Apostolos Vantarakis
Zitieren
APA
Palaiologou, A., Athanasiou, M., & Vantarakis, A. (2026). Concurrent Classification of Pseudomonas aeruginosa Occurrence From Routine Municipal Swimming‐Pool Monitoring Data: An Explainable Machine‐Learning Proof‐of‐Concept With Grouped Validation. Water Environment Research, 98(11). https://doi.org/10.1002/wer.70653
Vancouver
Palaiologou A, Athanasiou M, Vantarakis A. Concurrent Classification of Pseudomonas aeruginosa Occurrence From Routine Municipal Swimming‐Pool Monitoring Data: An Explainable Machine‐Learning Proof‐of‐Concept With Grouped Validation. Water Environment Research. 2026;98(11). doi:10.1002/wer.70653
BibTeX
@article{Palaiologou2026,
author = {Palaiologou, Athanasia and Athanasiou, Maria and Vantarakis, Apostolos},
title = {Concurrent Classification of Pseudomonas aeruginosa Occurrence From Routine Municipal Swimming‐Pool Monitoring Data: An Explainable Machine‐Learning Proof‐of‐Concept With Grouped Validation},
journal = {Water Environment Research},
year = {2026},
volume = {98},
number = {11},
doi = {10.1002/wer.70653}
}
EUVIMED