Feature Extraction And Pattern Classification Based On Multivariate Time Series

Heart disease,hypertension as a chronic disease,the characteristics of long duration,the characteristics of the cured,because people do not pay attention to in ordinary life,easy to bring great hidden life.Second,for a short medical instrument measuring heart,sometimes found no mutation of abnormal ecg signal,therefore,uses the telemedicine technology,by wearing long time monitoring of heart,blood pressure instrument can be abnormal ecg signal in time,have the effect of early found early prevention.However,for the long time signal collected by ECG and blood pressure detection instrument,huge amount of data will be generated and the burden of storage devices will increase.This will bring negative factors to the miniaturization of the ECG and blood pressure detector,and it is not conducive to the rapid diagnosis and treatment of doctors.Therefore,this paper mainly studies the dimensionality reduction of time series data and the classification of feature data based on the high-dimensional Dynamic Multivariate Time Series detected in telemedicine.First,due to the large amount of time series data and the noise interference,there are lots of redundant data.This part of data is very small for medical diagnosis.Therefore,we need to reduce the dimension and denoise the original ECG and blood pressure signals.Aiming at the problem that multivariate time series can not be reduced at the same time,2DPCA and 2DPCA are two dimensional principal component analysis methods.They are mainly used for image processing.Compared with PCA,2DPCA is based on 2-D matrix instead of one-dimensional vector.So when processing data matrix,it does not need to tran sform it into a vector between feature extraction.When PCA is dealing with similar problems,the two-dimensional data matrix must be transformed into one dimensional vector.When dealing with Covariance matrix,the size of the 2DPCA is much smaller.It is used to quickly excavate useful data features and reduce the dimension of time series.The method is applied to data collection of ECG and blood pressure of 15 college students.The experimental results show that the performance evaluation index of two-dimensional principal component analysis is significant,and it meets the requirements of multivariate high-dimensional time-series signal feature dimension reduction.Secondly,in order to achieve the classification of different pathological features of ECG and blood pressure detection signal,this paper proposes a decision tree with an optimized C4.5 classification method,genetic algorithm for optimization of decision tree,we will be the original decision tree T in all edges followed by top-down numbers,then for each number of edges and we give it a 0/1 weight.In this way,we convert the tree structure of the decision tree to the set of edges.Under the premise of never reducing the classification accuracy of decision tree,the size of the tree is reduced by setting appropriate fitness function.The more than 5000 sample is applied to the U.S.medical database wrist 20 data file 4 data processing conditions,and the data will be used in the decision tree classification method,the comparison of the test results show that the performance of the decision tree method based on genetic algorithm can effectively improve the decision tree in terms of classification accuracy and scale control etc.Finally,combined with the laboratory ECG and blood pressure detector,a dynamic multiple physiological signal sampling experiment was designed.The experiment was divided into two parts:quiet state electrocardiogram(ECG),blood pressure detection,motor state ECG and blood pressure detection.Firstly,the wavelet filter for ECG signal acquisition,single cycle segmentation,and then the matrix space 2DPCA based on dimension reduction method is applied to ECG and blood pressure detection signal dimension and characteristic data after dimensionality reduction using feature classification decision tree classification method based on genetic algorithm,verified the performance of decision tree method based on genetic algorithm improved decision tree in terms of classification accuracy and scale control.