The migraine is a common disease with high incidence. It is still not easy to explain its pathogenesis very well. Therefore, it lacks effective diagnostic methods. This study aims to predict migraine by using the functional magnetic resonance imaging technology to obtain functional network of brain, then through deep learning of automatically it extracts data features by Autoencoder, combined with various machine learning algorithms to provide a reference for clinical diagnosis of physicians. It can get better classification effect to extract data features and train the classifier by deep learning. The deep learning algorithm, based on the initial features obtained by the traditional templates, can further extract more fine and effective features, and obtain better classification performance in predicting migraine.