A Study With Brain Age Based On Feature Extraction And Machine Learning Algorithms

The brain has a functional decline as the age increases.This article explores the differences in the characteristics of the brain at different ages,and for this reason,the 18 participants were selected to make our decision-making(DM)experiment,in which the eight young(20-30 years)subjects and the eight elderly(older than 45 years)subjects.During the DM experiment,the brain wave(EEG)signals of the participants'responses of decision were fully recorded.First of all,in order to distinguish the charcteristics of EEG signals between young and the elderly,two different feature extraction methods of EEG signals were proposed in this paper,namely,the feature extraction of beta wave and the improved method of feature extraction of power spectral entropy(PSE)based on beta wave.It is indicated that the imporved feature extraction method can better characterize the brain wave with age characteristics in young and the elderly.Furthermore,Compared to delta wave,theta wave and alpha wave,the beta wave plays a leading role in the DM process of young and the elderly based on the feature extraction method of beta wave.Meanwhile,the PSE distribution of young was significantly larger than the elderly with the improved method of feature extraction.It is suggested that the brains of young were more informative than the elderly in the DM process.What's more,several machine learning classification algorithms were being to further differentiate the brain wave signals of young and the elderly,which were support vector machines(SVM),Random Forest(RF)and Extreme Learning Machine based Kernel(KELM)for the classification of beta datasets,and Logistic regression(LR),SVM,RF and XGBoost for the classification of PSE datasets.Firstly,the results are based on the beta datasets demonstrated that the KELM model performed better than the other two models and the highest accuracy reached 81.08%.In addition,the feature importance of the beta datasets were computed by the Gini index method and the C3 feature played a crucial role in the classification process,which is in the center of brain.In the C3 characteristic,the model was easier to divide the EEG signals of young and the elderly.Secondly,the results are based on the PSE datasets revealed that the XGBoost model performed well to the other three models and the highest accuracy was 91%,which is higher than the beta datasets in 9.92%.Therefore,the PSE feature extraction can well characterize the EEG signals between young and the elderly rather than the beta wave,and the good generalization and the robustness of the XGBoost model were in distinguishing brain age signals.In addition,the gain index of XGBoost was being to rank the feature importance with the PSE datasets,and the brain feature of the young and elderly people were the most different in the C3.Last but not least,the top six most important features of two datasets were chosen to infer the results that the differences of the brain wave of young and the elderly were most pronounced in the central and temporal regions of the brain.Therefore,the PSE feature extraction method and XGBoost model are more suitable for brain age diagnosis.By using machine learning algorithms the detect brain data,we can analyze the age of the elderly and prevent brain sub-health.Moreover,through the PSE analysis of EEG data in the brain,the elderly brains need more input to delay brain aging.