Logistic Discriminant Analysis Of Stroke Type And Recurrence And SMOTE Algorithm To Mine In Unbalanced Clinical Data

Objective:1Build a discriminant analysis model for stroke type diagnosis and strokerecurrence.2Compare SMOTE algorithm with random over sampling and subsampling method when dealing unbalanced data.Methods: Dealing medical cases of stroke patients and their data of healthmonitoring with SPSS and R. Logistic regression by SPSS Forward入=0.05，出=0.10Loading the package “DmWR” in R language to run SMOTEalgorithm when dealing with unbalanced data.Results:1Get a high correct percentage classification model of stroke patients.2Get a prediction model of stroke recurrence. Result of Cox&Snell RSquare test is0.634. Prediction accuracy is86.1%.3Get the prediction model of stroke recurrence by3kinds of resamplingmethod. Compared them by prediction accuracy and ROC.Conclusions:1The classification model can be used for assisting diagnosis.2Stroke Recurrence can be predicted by using the prediction model.3SMOTE algorithm is a better choice when dealing with unbalanced data.