In order to solve the problem that the traditional fMRI classification method has a low accuracy, this paper proposes a method that uses frequent subgraph mining and discriminative feature selection of weighted graphs for classification research based on independent component analysis and high-order functional connection network.This method does not need to rely on apriori brain tap template, but fully considers the time-varying characteristics in the scan time and applies frequent subgraph mining to the weighted map.At the same time, in order to more accurately find discriminative subgraph features, this paper also proposes several new discriminative feature selection methods.The results show that resting-state functional magnetic resonance imaging classification based on independent component and weighted high-order brain networks effectively improved the diagnostic accuracy of Alzheimer's disease.