Abstract:

Objective To evaluate the roles of random forest，extreme gradient boosting，LightGBM machine learning algorithms and Logistic regression analysis in predicting the prognosis of patients with acute ischemic stroke（AIS）. Methods The demographic and clinical data of 670 AIS patients from Fudan University Pudong Medical Center from February 2022 to June 2024 were collected. The 670 patients were randomly divided into a training set（469 cases） and a test set（201 cases） at a ratio of 7∶3. Totally，4 prediction models for AIS prognosis were established using Logistic regression analysis，random forest，extreme gradient boosting and LightGBM machine learning algorithm. The prognosis predictive effects were evaluated based on receiver operating characteristic（ROC） curve，accuracy，sensitivity，specificity，F1 score，positive predictive value and negative predictive value. Results The areas under curves（AUC） of Logistic regression analysis，random forest，extreme gradient boosting and LightGBM machine learning algorithms for predicting the prognosis of AIS patients were 0.908，0.915，0.908 and 0.896，respectively. Among the 4 algorithms，the random forest algorithm had the optimal accuracy and F1 score，Logistic regression analysis had the highest specificity and positive predictive value，and LightGBM machine learning algorithm had the optimal negative predictive value. The extreme gradient boosting algorithm performed poorly in all evaluation indicators. Conclusions The AIS prognosis prediction algorithms constructed based on machine learning can early identify the prognosis of patients，but an appropriate machine learning model should be selected.

Key words: Machine learning, Acute ischemic stroke, Prognosis evaluation model