Abstract: In order to solve the problem of classification and accuracy of epilepsy EEG( Electroencephalogram)signals，frequency slice wavelet transform was used to reconstruct EEG data and get five frequency bands of rhythmic signals． We use approximate entropy of non-linear index and fluctuation index of linear index as the eigenvalues of epileptic signals to fully extract the characteristic information of signals． Gradient lifting tree algorithm was used to classify the feature data set． The classification recognition rate of epileptic EEG signals is 98． 4%． Compared to the traditional Adaboost algorithm，we adopt GBDT( Gradient Boosting Decision Tree) as a classification algorithm． This method can use more data sets successfully and has higher classification accuracy．

Key words: epileptic electroencephalogram signal, frequency slice wavelet transform, approximate entropy, fluctuation index, gradient boosting decision tree