Fig. 5

Random forest machine learning models were constructed to classify stroke-attack patients and non-stroke controls. Using the threshold of stroke risk indicators as learning parameters to distinguish (A) fist-onset stroke patients and non-stroke controls, D recurrent stroke patients and non-stroke controls, G fist-onset stroke patients and recurrent stroke patients. Using both of stroke risk indicators and PAGln level as learning parameters to distinguish (B) fist-onset stroke patients and non-stroke controls, E recurrent stroke patients and non-stroke controls, H fist-onset stroke patients and recurrent stroke patients. C, F, I Variable importance diagram