Epilepsy Detection Based On Wavelet Analysis And GBDT Algorithm

Epilepsy is a chronic disease that causes transient brain dysfunction caused by unconventional discharge of brain neurons.According to a report by the World Health Organization in 2005,the most common psychiatric disease is epilepsy.At present,Electroencephalogram(EEG)is an important method for assessing brain activity,which is called the “gold standard” for the diagnosis of epilepsy.However,doctors visually examine the patient's EEG to achieve a clinical diagnosis of epilepsy,but this method is inefficient and easily leads to misdiagnosis.Therefore,the use of the current popular pattern recognition technology and computer technology to deal with EEG signals has become an important diagnostic tool for epilepsy.In this paper,the automatic diagnosis algorithm is complex and the classification accuracy is not high.The classification algorithm is improved.A method based on Gradient Boosting Decision Tree(GBDT)algorithm for epilepsy detection is proposed and compared with the classical support vector machine algorithm.Comparison,the main work is as follows:1.Through analysis of long-term EEG characteristics,it is found that epileptic EEG and intermittent EEG have large differences in amplitude,frequency,and complexity.By preprocessing and feature extraction of EEG signals,classifiers can be used.Quickly and effectively distinguish between the two.In the stage of EEG signal processing,the feature extraction based directly on EEG signals will cause a lot of details to be lost.In this paper,the frequency slice wavelet transform is used to reconstruct the EEG data,and the rhythm signals of five frequency bands are obtained.The linear index approximate entropy and the linear index fluctuation index are used together as the eigenvalues of the epileptic signal,and the characteristic information of the signal is fully extracted;2.In order to improve the effectiveness of the classification effect,this paper compares the classical method support vector machine classifier with the GBDT classifier,and uses the Genetic Algorithm(GA)to optimize the optimal set of punishments among many possible initial values.The factor and sum function parameters were combined with the Support Vector Machine(SVM)for simulation testing.The EEG feature vector is sent to the trained classifier to classify the EEG signals to distinguish between Epileptic EEG and intermittent EEG.Thereby achieving detection of seizures.In comparison with the classical method support vector machine classifier and GBDT classifier,the classification rate of epilepsy EEG signals is 98.4%.The experimental results show that the GBDT classifier successfully uses more data sets and calculates the speed.The advantages of fast and high classification accuracy are more suitable for clinical applications.