A comprehensive study based on machine learning models for early identification Mycoplasma pneumoniae infection in segmental/lobar pneumonia.

Segmental/lobar pneumonia in children following Mycoplasma pneumoniae (MP) infection has a significant threat to the children's health, so early recognition of MP infection is critical to reduce the severity and improve the prognosis of segmental/lobar pneumonia in children. In this study, we aim to build predictive models using machine learning techniques to assist clinicians in the early identification of MP infection. We collected medical records of children with segmental/lobar pneumonia at the First Hospital of Jilin University between December 2016 and December 2021, and used four machine learning models for testing and validation. In this study, a total of 630 cases of children with segmental/lobar pneumonia were collected. After data pre-processing and feature selection, seven variables were used for prediction model construction. Four machine learning models were applied to predictive learning, and selecting Random Forest as a prediction model for MP infection after comprehensive selection, which achieved 57.1% sensitivity, 69.6% accuracy and 0.752 AUC. Based on machine learning algorithms, combined with conventional indicators of segmental/lobar pneumonia in children, to construct a prediction model for early identification of MP infection, which is of great help in assisting clinicians in early, targeted treatment.