Exploring machine learning methods for predicting systemic lupus erythematosus with herpes

Abstract Objectives To investigate whether machine learning, which is widely used in disease prediction and diagnosis based on demographic data and serological markers, can predict herpes occurrence in patients with systemic lupus erythematosus (SLE). Methods A total of 286 SLE patients were included in this study, including 200 SLE patients without herpes and 86 SLE patients with herpes. SLE patients were randomly divided into a training group and a test group, and 18 demographic characteristics and serological indicators were compared between the two groups. Results We selected basophil, monocyte, white blood cell, age, immunoglobulin E, SLE Disease Activity Index, complement 4, neutrophil, and immunoglobulin G as the basic features of modeling. A random forest model had the best performance, but logistic and decision tree analyses had better clinical decision‐making benefits. Random forest had a good consistency between feature importance judgment and feature selection. The 10‐fold cross‐validation showed the optimization of five model parameters. Conclusion The random forest model may be an excellently performing model, which may help clinicians to identify SLE patients whose disease is complicated by herpes early.