Machine learning-assisted prediction of pneumonia based on non-invasive measures

• DOI: 10.3389/fpubh.2022.938801
• Publication Date: 2022-07-28T10:12:26Z
• Authors: Clement Yaw Effah; Ruoqi Miao; Emmanuel Kwateng ...; Clement Agboyibor; Ruiping Qiao; Yongjun Wu; Lijun Miao; Yanbin Wang
• Keywords: non-invasive measures; Pneumonia; 3. Good health; Machine Learning; 03 medical and health sciences; machine learning; 0302 clinical medicine; electronic health records (EHR); pneumonia; Humans; Public Health; Public aspects of medicine; RA1-1270; decision support system (DSS)

Pneumonia is an infection of the lungs that characterized by high morbidity and mortality. The use machine learning systems to detect respiratory diseases via non-invasive measures such as physical laboratory parameters gaining momentum has been proposed decrease diagnostic uncertainty associated with bacterial pneumonia. Herein, this study conducted several experiments using eight models predict pneumonia based on biomarkers, parameters, features. We perform machine-learning analysis 535 different patients, each 45 Data normalization rescale all real-valued features was performed. Since it a binary problem, we categorized patient into one class at time. designed three evaluate models: (1) feature selection techniques select appropriate for models, (2) imbalanced original dataset, (3) SMOTE data. then compared their effectiveness in predicting Biomarkers C-reactive protein procalcitonin demonstrated most significant discriminating power. Ensemble RF (accuracy = 92.0%, precision 91.3%, recall 96.0%, f1-Score 93.6%) XGBoost 90.8%, 92.6%, 92.3%, f1-score 92.4%) achieved highest performance accuracy dataset AUCs 0.96 0.97, respectively. On prediction results f1-scores 92.0 91.2%, Also, AUC 0.97 both models. Our showed diagnosis pneumonia, individual clinical history, indicators, symptoms do not have adequate discriminatory can also conclude ensemble ML performed better study.