A machine learning framework for identifying influenza pneumonia from bacterial pneumonia for medical decision making.

Accurately identifying viruses from bacteria determine the therapy and outcome of patients suffering from pneumonia. This study focuses on exploring the capability of machine learning techniques in detecting the well-known influenza virus pneumonia from bacterial pneumonia based on the key variable quantity of blood tests and computer tomography (CT). In this study, we propose an improved Harris hawk's optimization (GSHHO) based on the Gaussian mutation mechanism and simulated annealing method, and then combine GSHHO with Fuzzy k-nearest neighbor (FKNN) to form a GSHHO-FKNN machine learning model, GSHHO-FKNN. The experimental results have demonstrated that the proposed GSHHO-FKNN has identified that ground grass opacity (GGO) and bronchovascular bundle were distinctive CT phenomenon of influenza pneumonia. Furthermore, the number of procalcitonin (PCT) and white blood cells (WBC) in the cases of influenza pneumonia were notable from those in the cases of bacterial pneumonia. In addition, an interesting result was that the absolute value of eosinophils, platelet and flake shadow in influenza pneumonia cases was more abnormal than those of bacterial pneumonia cases. It proved that the GSHHO-FKNN model was an effective method for discriminating influenza pneumonia from community acquired pneumonia, leading to accurate diagnosis against pathogenic microorganisms. • We firstly propose an improved Harris hawks optimization (GSHHO). • Experiments are carried out on the benchmark functions to verify the performance of the proposed algorithm GSHHO. • Combine GSHHO with Fuzzy k-nearest neighbor (FKNN) to form a GSHHO-FKNN machine learning model. • The proposed GSHHO-FKNN model aims to maximize the generalization capability of the FKNN classifier. • It is well proved that this is an effective machine learning framework (GSHHO-FKNN). [ABSTRACT FROM AUTHOR]