The physical condition of drivers is closely related to traffic safety, especially the driver’s cardiovascular health condition. Real-time monitoring of drivers’ health can help drivers understand their physical condition in time and reduce traffic accidents caused by sudden illnesses. In this study, firstly, 657 PPG (Photoplethysmography Signal) pulse wave datasets from Guilin People’s Hospital, Guangxi Zhuang Autonomous Region, China, were dichotomized numerically for cerebrovascular diseases after the noise reduction by Chebyshev Ⅱ filter and the extraction of time domain features, frequency domain features and wavelet packet features by fast Fourier method. Then, the numerically labeled cerebrovascular disease types were used as output parameters to construct driver cerebrovascular disease dataset. To solve the problem of unbalanced classification of samples in actual dataset, an oversampling supplement was performed by the SMOTE algorithm and a driver cerebrovascular disease classification model, namely SSA-DELM, was constructed based on PPG feature values, followed with model training and testing on actual datasets. The results show that the proposed classification model can provide comparatively accurate early warning for drivers suffering from cerebral infarction or cerebrovascular disease, with an accuracy of 83%, an average precision of 80%, a completeness of 76.6%, an F₁ score of 0.79, and a mean average precision of 0.80. This research can provide theoretical model basis and technical support for drivers’ dynamic health monitoring system based on PPG signal. The proposed model has a large application space in the software service and intelligent medical care of the new energy automobile industry, which is in line with the sales mode of the whole industry chain of “terminal + software + service” of new energy automobile enterprises, and is also in line with modern people’s attention to environmental protection, family health and intelligent transportation.