Optimal machine learning model selection and validation for predicting emergency response capability of stroke caregivers in stroke recurrence
LI Bo
YANG Mingying
WANG Ya
ZHAN Anning
XIAO Yang
YANG Qiqi
YANG Zihang
Abstract：Objective To apply machine learning algorithms to construct a stroke recurrence emergency response ca-pability prediction model for stroke caregivers and to select and verify the model with the best predictive performance.Methods A total of 515 caregivers of stroke patients hospitalized from April to August 2024 were recruited.Caregivers were categorized into"deficient"and"non-deficient"groups based on their emergency response capacity deficits.Four ma-chine learning algorithms,random forest,artificial neural network,extreme gradient boosting,and gradient boosting deci-sion tree(GBDT),were employed to construct predictive models.The performance of the models was compared using ac-curacy,precision,recall,specificity,sensitivity,Youden's index,and the area under the receiver operating characteris-tic curve(AUC).The Gini index was applied to determine significant influencing factors of stroke caregivers'emergency response capacity.Results The GBDT model demonstrated the best performance in predicting stroke recurrence emer-gency response capability among stroke caregivers,with an AUC value reaching 0.896,indicating high prediction accura-cy.Ten core predictors were identified in the GBDT model,including caregiver burden score,social support score,educa-tion level,personal burden score,relationship to the patient,age,primary family economic provider status,participation in stroke knowledge education programs,monthly income,and subjective support score.Conclusions By comparing multiple machine learning algorithms,this study found that the GBDT model excelled in predicting stroke recurrence emer-gency response capability of stroke caregivers.The model effectively pinpointed critical factors influencing this capacity,enabling dynamic monitoring of predictor changes.These findings lay a technical foundation for personalized intervention protocols,thereby forming a closed-loop support system to improve home-based care quality for stroke patients.
Keywords：machine learningstrokecaregiversemergency response capabilityprediction model
Publication Date：2025-04-25
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：6( 1-6 )