Prediction of COVID-19 disease severity using synthetic data oversampling and machine learning methods on data at first hospitalization

Kübra Köksal; Buket Doğan; Zehra Aysun Altıkardeş

Makale detayları

Prediction of COVID-19 disease severity using synthetic data oversampling and machine learning methods on data at first hospitalization

Yazarlar:

Kübra Köksal | Buket Doğan | Zehra Aysun Altıkardeş

Yayıncı:

Gazi University

Ana Başlıklar:

Makine Öğrenme (Diğer)

Dergi:

Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.40 Issue 1, pp.413-428

Belge Tipi:

Makale

Yayın Tarihi:

8/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001317623

pISSN:

1300-1884

Anahtar kelimeler:

Özet:

 Covid-19 is an infectious disease caused by the SARS-CoV-2 virus, which first appeared in Wuhan, China in December 2019 and was declared a pandemic by the World Health Organization on March 11, 2020. Since its first appearance, Covid-19 has spread rapidly around the world, causing the start of a process that will negatively affect all human life, especially the health sector. In addition to measures such as hygiene, mask, distance and vaccine in the fight against Covid-19, it is aimed to provide benefits in Covid-19 diagnosis and prediction processes by developing computer-aided systems by researchers. In this study, which was developed in this direction, it is aimed to develop and test machine learning models that help predict WHO-oriented disease severity by using the laboratory and demographic characteristics of patients infected with COVID-19 at the hospital admission stage. In the study, a domestic data set created by using the information obtained from the patients who applied to Marmara University Hospital was used. The relationship between two different endpoints, namely oxygen need and intensive care need, and the first laboratory results on the dataset was analyzed using K-Nearest Neighbor, Bagging, Random Forest and Decision Tree machine learning methods. The unbalanced class distribution in the data set was balanced using the SMOTE data multiplication algorithm, and the effect of data multiplication on classification performance was evaluated in terms of accuracy and F1-Score. On the data set without SMOTE, the oxygen requirement of the patient during the first hospitalization was estimated with 16 features with 91.67 accuracy, the oxygen requirement during hospitalization with 18 features with 91.96, and the intensive care need at hospitalization with 12 features with an accuracy of 92.17. After SMOTE data multiplication, an increase of 6 for Analysis-1, 23 for Analysis-2 and 21 for Analysis-3 was observed in the F1-Score values of minority classes.

 Covid-19 is an infectious disease caused by the SARS-CoV-2 virus, which first appeared in Wuhan, China in December 2019 and was declared a pandemic by the World Health Organization on March 11, 2020. Since its first appearance, Covid-19 has spread rapidly around the world, causing the start of a process that will negatively affect all human life, especially the health sector. In addition to measures such as hygiene, mask, distance and vaccine in the fight against Covid-19, it is aimed to provide benefits in Covid-19 diagnosis and prediction processes by developing computer-aided systems by researchers. In this study, which was developed in this direction, it is aimed to develop and test machine learning models that help predict WHO-oriented disease severity by using the laboratory and demographic characteristics of patients infected with COVID-19 at the hospital admission stage. In the study, a domestic data set created by using the information obtained from the patients who applied to Marmara University Hospital was used. The relationship between two different endpoints, namely oxygen need and intensive care need, and the first laboratory results on the dataset was analyzed using K-Nearest Neighbor, Bagging, Random Forest and Decision Tree machine learning methods. The unbalanced class distribution in the data set was balanced using the SMOTE data multiplication algorithm, and the effect of data multiplication on classification performance was evaluated in terms of accuracy and F1-Score. On the data set without SMOTE, the oxygen requirement of the patient during the first hospitalization was estimated with 16 features with 91.67 accuracy, the oxygen requirement during hospitalization with 18 features with 91.96, and the intensive care need at hospitalization with 12 features with an accuracy of 92.17. After SMOTE data multiplication, an increase of 6 for Analysis-1, 23 for Analysis-2 and 21 for Analysis-3 was observed in the F1-Score values of minority classes.

 Covid-19 is an infectious disease caused by the SARS-CoV-2 virus, which first appeared in Wuhan, China in December 2019 and was declared a pandemic by the World Health Organization on March 11, 2020. Since its first appearance, Covid-19 has spread rapidly around the world, causing the start of a process that will negatively affect all human life, especially the health sector. In addition to measures such as hygiene, mask, distance and vaccine in the fight against Covid-19, it is aimed to provide benefits in Covid-19 diagnosis and prediction processes by developing computer-aided systems by researchers. In this study, which was developed in this direction, it is aimed to develop and test machine learning models that help predict WHO-oriented disease severity by using the laboratory and demographic characteristics of patients infected with COVID-19 at the hospital admission stage. In the study, a domestic data set created by using the information obtained from the patients who applied to Marmara University Hospital was used. The relationship between two different endpoints, namely oxygen need and intensive care need, and the first laboratory results on the dataset was analyzed using K-Nearest Neighbor, Bagging, Random Forest and Decision Tree machine learning methods. The unbalanced class distribution in the data set was balanced using the SMOTE data multiplication algorithm, and the effect of data multiplication on classification performance was evaluated in terms of accuracy and F1-Score. On the data set without SMOTE, the oxygen requirement of the patient during the first hospitalization was estimated with 16 features with 91.67 accuracy, the oxygen requirement during hospitalization with 18 features with 91.96, and the intensive care need at hospitalization with 12 features with an accuracy of 92.17. After SMOTE data multiplication, an increase of 6 for Analysis-1, 23 for Analysis-2 and 21 for Analysis-3 was observed in the F1-Score values of minority classes.

 Covid-19 ilk olarak 2019 Aralık ayında Çinin Wuhan kentinde ortaya çıkan ve 11 Mart 2020de Dünya Sağlık Örgütü tarafından pandemi olarak ilan edilen, SARS-CoV-2 virüsünün neden olduğu bir bulaşıcı hastalıktır. Covid-19 ilk ortaya çıktığı tarihten itibaren dünya genelinde hızla yayılarak başta sağlık sektörü olmak üzere tüm insan hayatını olumsuz yönde etkileyecek bir sürecin başlamasına neden olmuştur. Covid-19 ile mücadelede hijyen, maske, mesafe ve aşı gibi önlemlerin yanında araştırmacılar tarafından bilgisayar destekli sistemler geliştirilerek Covid-19 teşhis ve tahmin süreçlerinde fayda sağlanması hedeflenmiştir. Bu doğrultuda geliştirilen bu çalışmada COVID-19 ile enfekte olmuş hastaların hastaneye kabul aşamasında alınan laboratuvar ve demografik özellikleri kullanılarak WHO odaklı hastalık şiddetini tahmin etmeye yardımcı olan makine öğrenmesi modellerinin geliştirilmesi ve test edilmesi amaçlanmaktadır. Çalışmada Marmara Üniversitesi Hastanesine başvuran hastalardan alınan bilgiler kullanılarak oluşturulan yerli bir veri seti kullanılmıştır. Veri seti üzerinde oksijen ihtiyacı ve yoğun bakım ihtiyacı olmak üzere belirlenen iki farklı sonlanım durumu ile ilk laboratuvar sonuçları arasındaki ilişki K-En yakın komşu, Torbalama Bagging, Rastgele Orman ve Karar Ağacı makine öğrenmesi yöntemleri kullanılarak analiz edilmiştir. Veri setindeki dengesiz sınıf dağılımı SMOTE veri çoğaltma algoritması kullanılarak dengeli bir hale getirilmiş ve veri çoğaltmanın sınıflandırma performansına etkisi doğruluk ve F1-Skor açısından değerlendirilmiştir. SMOTE uygulanmayan veri seti üzerinde hastanın ilk yatış aşamasındaki oksijen ihtiyacı Analiz  1 16 özellik ile 91,67, yatış sırasındaki oksijen ihtiyacı Analiz  2 18 özellik ile 91,96 ve yatış sırasındaki yoğun bakım ihtiyacı Analiz  3 12 özellik ile 92,17 doğruluk değeri ile tahmin edilmiştir. SMOTE veri çoğaltma işleminden sonra azınlık sınıfların F1-Skor değerlerinde Analiz  1 için 6lık, Analiz  2 için 23lik ve Analiz  3 21lik bir artış gözlenmiştir.

Makale detayları

Prediction of COVID-19 disease severity using synthetic data oversampling and machine learning methods on data at first hospitalization

Yazarlar:

Kübra Köksal | Buket Doğan | Zehra Aysun Altıkardeş

Yayıncı:

Gazi University

Ana Başlıklar:

Makine Öğrenme (Diğer)

Dergi:

Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.40 Issue 1, pp.413-428

Belge Tipi:

Makale

Yayın Tarihi:

8/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001317623

pISSN:

1300-1884

Covid-19

machine learning

laboratory data

prognosisCovid-19

makine öğrenmesi

laboratuvar verisi

prognoz

Sayı

Makale detayları

Yazarlar: Kübra Köksal | Buket Doğan | Zehra Aysun Altıkardeş

Yayıncı: Gazi University

Ana Başlıklar: Makine Öğrenme (Diğer)

Dergi: Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.40 Issue 1, pp.413-428

Belge Tipi: Makale

Yayın Tarihi: 8/2024

Belge Dili: Türkçe

Document ID (Org No.): 2001317623

pISSN: 1300-1884

Sayı

Yazarlar:

Kübra Köksal | Buket Doğan | Zehra Aysun Altıkardeş

Yayıncı:

Gazi University

Ana Başlıklar:

Makine Öğrenme (Diğer)

Dergi:

Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.40 Issue 1, pp.413-428

Belge Tipi:

Makale

Yayın Tarihi:

8/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001317623

pISSN:

1300-1884