Research On Pattern Recognition Of EEG Based On Wavelet Packet

In recent years,research on brain waves has become more and more intense.With the development of data mining and artificial intelligence,the use of brain wave data has become more and more in-depth.Nowadays brain waves can help doctors diagnose brain diseases in the medical field.They can also make machine controled by brain waves to help facilitate people with disabilities.They can also use brain waves as lie detectors to interrogate prisoners.This article uses brain waves to identify whether the human eye is blinking or closed.In order to better understand the whole data,this paper first describes the statistical analysis of the data,and based on this eliminates the outliers in the data,the outliers have a great influence on the results of the model,so we remove it directly.In order to observe the entire data set more intuitively,we performed visual processing on the data,including exploring the timing diagram of each variable and the correlation between variables.At the same time,we used TSNE and PCA to reduce the dimension of the data and then visualize it.Then,we use six classification algorithms including SVM,KNN,decision tree,random forest,GBDT,and XGBoost to classify,and use CV cross-validation to perform parameter optimization for each model.Brain wave data belongs to a type of signal data,and signals in real are often accompanied by noise.This paper will use wavelet denoising to process data,and comparing the data before and after processing has no significant effect on the model results.The results show that the noise contained in the data has a certain impact on the model results.The denoised data training model and the model results have been significantly improved.Here,we use AUC as the main discriminant index.In the above model,we compare the model results.The best performing model is SVM,The AUC value is as high as 0.9924,followed by KNN,and the AUC value is as high as 0.9900.At the same time,this paper discusses the improvement of the model's results by model fusion.The three models of stacking,blending,and voting are used respectively.The results show that the stacking model has the best fusion effect.