Naslov
Analysis of COVID-19 disease using machine learning - personalized model

Autori
Blagojević, Anđela ; Šušteršič, Tijana ; Lorencin, Ivan ; Baressi Šegota, Sandi ; Anđelić, Nikola ; Milovanović, Dragan ; Baskić, Dejan ; Car, Zlatan ; Filipović, Nenad

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
1st Serbian International Conference on Applied Artificial Intelligence (SICAAI) - book of abstracts / Filipović, Nenad - Kragujevac : Univerzitet u Kragujevcu, 2022

ISBN
978-86-81037-71-3

Skup
1st Serbian International Conference on Applied Artificial Intelligence (SICAAI)

Mjesto i datum
Kragujevac, Srbija, 19.05.2022. - 20.05.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
COVID-19 ; personalized model ; clinical condition assessment ; ensemble model ; rule-based machine learning

Sažetak
Coronavirus disease (COVID-19), since its appearance, has put a large burden on the global health system which have strived to mitigate the pandemic, but mortality of COVID-19 continues to increase. Many authors have employed machine learning (ML) algorithms in the investigation of COVID-19 in order to identify infected individuals, predict their condition in time, predict the outbreaks and forecast certain numbers. Although there are many studies that examine the application of ML in the diagnosis of prognostic biomarkers and survival prediction several days in advance, there is a limited literature dealing with evidence to label patients in more categories (mild, moderate, severe, etc.) that would help not only to respond in a timely manner to prevent lethal results, but also to minimize the number of patients in hospitals where this is not the case. In this paper we present a methodology for classification of patients into 3 distinct classes of clinical condition (mild, moderate and severe) of COVID-19 disease and prediction of the outcome (change of severity of clinical condition) in advance. The results show that XGBoost classifier achieved average accuracy of 88%. The main advantage of our system is that it is a rule-based algorithm which is easier to implement in a real clinical practice, instead of the use of black box models, which are not appealing for real clinical use.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA