Brusellozlu Hastalarda Bakteriyeminin Makine Öğrenmesi Yöntemleri Kullanılarak Tahmin Edilmesi

Abstract

Purpose: The correct and early diagnosis of brucellosis is very crucial to decelerate its spread and providing fast treatment to patients. This study aims to develop a predictive model for diagnosing bacteremia in brucellosis patients based on some hematological and biochemical markers without the need for blood culture and bone marrow and to investigate the importance of these markers in predicting bacteremia. Materials/Methods: 162 patients with diagnosing brucellosis, 54.9% of whom are non-bacteremic, 45.1% bacteremia were retrospectively collected. The 20 demographic, hematological and biochemical laboratory parameters and 30 classifiers are used to predict bacteremia in brucellosis. Classifiers were developed by using Python programming language. Accuracy (ACC), Area under the receiver operating characteristic curve (AROC), and F measure were employed to find the best fit classification method. Feature importance method was used to determine most diagnostic markers to predict the bacteremia. Results: Extratree classifier with criterion “entropy” (ETC1) showed the best predictive performance with Acc values ranging between 0.5 and 1.00, F values between 0.53 and 1, and AROC values between 0.62 and 1. The neutrophil%, lymphocyte%, eosinophil%, alanine aminotransferase, and C-reactive protein were determined as the most distinguishing features with the scores 0.723, 1.000, 0.920, 0.869, and 0.769, respectively. Conclusions: This study showed that the ETC1 classifier may be helpful in determining bacteremia in brucellosis patients and that elevated lymphocytes, alanine aminotransferase, and C-reactive protein and low neutrophils and eosinophils may indicate bacteremic brucellosis.