Prognostic Performance of Statistical and Machine Learning Methods on MIMIC-III Clinical Database

Özalp M. A., Yıldırak K., Aladağ C., Ünal M. N.

International Conference on Data Science, Machine Learning and Statistics - 2019 (DMS-2019), H. Eray Celik Cagdas Hakan Aladağ, Editör, Van Yüzüncü Yıl Üniversitesi, Van, ss.8-11, 2019

  • Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
  • Basım Tarihi: 2019
  • Yayınevi: Van Yüzüncü Yıl Üniversitesi
  • Basıldığı Şehir: Van
  • Sayfa Sayıları: ss.8-11
  • Editörler: H. Eray Celik Cagdas Hakan Aladağ, Editör
  • Ankara Üniversitesi Adresli: Evet

Özet

MIMIC-III is a data set composed of more than 60000 admissions made to Beth Israel Hospitals. For every deidentified
critical care patients demographics, vital signs, laboratory tests, medications, and more are hold in this
database. The most important cause of deaths in the hospital is considered as Sepsis. Sepsis is defined as ‘lifethreatening
organ dysfunction caused by a dysregulated host response to infections. In medical literature, many
scoring systems such as SOFA, LODS, SIRS, NEWS, etc. have been suggested for the early prediction/diagnosis of
sepsis and evaluation of prognosis. Both machine learning and statistical learning methods have been applied to
model survival/death status for intensive care unit patients in Mimic - III database. Used methods are Random Forest,
Support Vector Machine, Logistic Regression, Naive Bayes, Adaboost and Artificial Neural Networks (ANN). It is
a well-known fact that ANN approach is an effective prediction tool. And, it is very crucial to determine the best
ANN model in order to get accurate predictions. In this study, different ANN models have been applied to MIMICIII
data set to determine the best ANN model. As a result of the implementation, all obtained prognostic results are
presented and discussed.