As a common disease, migraine has a large patient base and a high incidence rate, which deeply affects the daily life of patients. With the continuous development of artificial intelligence technology, machine learning has developed rapidly and achieved good results in the fields of disease classification and prediction. Now, with the help of machine learning technology, a classification model for migraine symptoms was proposed. To address the issue of data imbalance, SMOTE (synthetic minority oversampling technique) was combined with the Tomek links method for data processing. Traditional machine learning methods and ensemble learning methods such as CatBoost(categorical boosting), LightGBM(lightweight gradient boosting machine), GBDT(gradient boosting decision trees), Stacking were used for training and testing. By using Bayesian optimization method for hyperparameter tuning and comparing the performance of multiple models, the final results show that the model using CatBoost, LightGBM, and GBDT as Stacking based learners performs the best, with a prediction accuracy of 98.84%. This discovery indicates that the model has excellent performance in migraine classification, and provides an effective method for improving the accuracy and reliability of migraine classification.