Relapse risk prediction for children with henoch-schonlein purpura based on GA-SVM

Abstract

The relapse risk prediction for children with Henoch-Schonlein purpura can help pediatricians make an accurate prognosis and offer personalized and appropriate follow-up nursing and relapse control to patients. In this study, we propose a Genetic algorithm-Support vector machine (GA-SVM) learning method combining the support vector machine with the genetic algorithm for parameter optimization to capture the nonlinear mapping from a panel of biomarkers to the relapse risk of HSP children. The GA-SVM prediction model is created by using the dataset of 40 samples in clinical treatment and observation of patients. The inputs of the model consist of 19 biomarkers including gender, age, immunoglobulin M, immunoglobulin G, immunoglobulin A, prothrombin time, etc. The outputs consist of 1 and-1, where 1 indicates high relapse risk and-1 indicates low relapse risk. For comparison, the GS-SVM prediction model based on parameter optimization of grid search is also created. The experimental results show that the GA-SVM prediction model has a high prediction accuracy of 90% and is strong in generalization ability. The GA-SVM model for predicting the relapse risk of HSP children is a promising decision support tool of clinical prognosis, which provides pediatricians with valuable assistance to offer rehabilitation treatment to patients.