Authors:
Emanuela Guglielmi
1
;
Davide Donato Russo
1
;
Pasquale Trinchese
1
;
Gennaro Laudato
1
;
Simone Scalabrino
1
;
Gianluca Testa
2
and
Rocco Oliveto
1
Affiliations:
1
Department of Bioscience and Territory, University of Molise, Italy
;
2
Department of Medicine and Health Sciences, University of Molise, Italy
Keyword(s):
Signal Noise, Machine Learning, Empirical Study, Atrial Fibrillation, Ventricular Tachycardia.
Abstract:
The Internet of Medical Things (IoMT) plays a vital role in healthcare by enhancing preventive careand chronic disease management through continuous monitoring using smart sensors and wearable devices. However, the reliability of IoMT systems can be compromised by noise in the acquired vital signals, which can negatively impact the accuracy of Machine Learning (ML) models used for anomaly detection. This study evaluates the impact of various disturbances on the performance of ML models in predicting cardiac conditions, with a focus on assessing the reliability and effectiveness of these systems in real-world applications. We investigated the effects of three types of noise—baseline wander, muscle artifact noise, and electrode motion artifact—on the performance of two advanced ML models designed to predict cardiac conditions, specifically atrial fibrillation (AF) and ventricular tachycardia (VT). Our analysis centered on how different noise intensities (i.e., the “loudness” of the noi
se) and durations (i.e., the length of time the noise persists) impacted the classification performance of these models. The VT detection model showed robust performance, with minimal impact even under intense and prolonged noise conditions. In contrast, AF detection was affected by all types of noise, with classification accuracy decreasing by up to ∼59% in the most challenging scenarios.
(More)