Information Fusion for Seveirty Detection using Machine Learning in Data Mining Healthcare

Improved patient care and less mental strain on healthcare providers are two benefits of using algorithms for machine learning to healthcare data. These algorithms may be used to spot irregularities in vital signs, which might speed up medical assistance or provide light on a disease's progress...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2023 International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) s. 1 - 6
Hlavní autoři: Gupta, Vaishali, Bhambu, Pawan
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 29.04.2023
Témata:
and Data Mining Healthcare
Data models
Distributed databases
Heart rate
Information Fusion
Machine learning
Machine learning algorithms
Medical services
Severity Detection
Training
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Improved patient care and less mental strain on healthcare providers are two benefits of using algorithms for machine learning to healthcare data. These algorithms may be used to spot irregularities in vital signs, which might speed up medical assistance or provide light on a disease's progression. While there is a wealth of literature comparing the unsupervised and supervised performances of anomaly detection algorithms on popular public datasets, this same level of conceptual comparison is lacking when it comes to physiological data. Knowing one's heart rate may provide valuable insight on one's health and level of physical activity, making it an underutilised data source. Specifically, we used and compared five machine learning methods, two of which were unsupervised and the other three supervised, to identify outliers in heart rate data. The algorithms were tested using physiological data from human subjects' hearts. Results demonstrated that both outlier factor and regression trees algorithms were effective in detecting heart rate anomalies, with both models successfully generalising from their simulation heart rate data training to real-world heart rate data. In addition, the findings lend credence to the idea that, in the absence of real labelled data, simulated data can be used to configure methodologies to a certain degree of performance, indicating that this kind of training could be particularly useful in the initial rollout of a system with no preexisting data.
DOI:10.1109/ICDCECE57866.2023.10151332