Research On ECG Anomaly Detection Base On Machine Learning Method

With the increase of people’s life pressure and work pressure,the prevalence of cardiovascular and cerebrovascular diseases increases year by year and shows the characteristics of younger age.Remote ECG monitoring can carry out 24-hour continuous monitoring of human heart activities,realize automatic analysis of ECG signals,early detection of abnormal ECG signals and early warning to patients.However,there are also some problems in the use of ECG monitor: including the falling off of electrode and the disconnection between lead wire and electrode in the process of monitoring,resulting in flake or disordered waveform of ECG and unable to recognize abnormal ECG signal;Inaccurate electrode placement or poor contact of electrodes and wires leads to baseline drift;The noise in the process of ECG signal acquisition and the incorrectly labeled samples in the training set lead to poor recognition accuracy.Among these problems,the main reason restricting the use of remote ECG monitoring is the poor classification accuracy caused by the incorrectly labeled samples in the training set.Therefore,it is of great significance for remote ECG monitoring to realize the recognition of incorrectly labeled samples in the training set and improve the classification accuracy of the classifier.Based on the above reasons,this paper implements a new method to identify the wrong labeled samples in ECG training set.On the basis of cross validation,this method combines iterative training and ensemble learning,uses ensemble learning to preliminarily identify the wrong labeled samples in the training set,and then further identifies the wrong labeled samples in the training set through iteration.While improving the recognition accuracy,it reduces the error recognition rate,and finally improves the classification performance of the classifier.In the aspect of ECG automatic classification,support vector machine has great advantages for the classification of small data samples,but the classification performance of support vector machine is closely related to the parameters of its kernel function and penalty coefficient.In view of the disadvantage that the standard particle swarm optimization algorithm is easy to fall into local optimization when looking for parameters.This paper studies the relationship between the inertia weight value and the search ability of particle swarm optimization algorithm,and puts forward the sine function inertia weight value,so that the particle swarm optimization algorithm can jump out of the local optimum when looking for parameters.The main work of this paper is as follows:The first chapter introduces the research background and significance of this paper.The development status of ECG anomaly detection and the existing problems in the field of ECG classification are reviewed.The second chapter first briefly describes the generation principle of ECG,the waveform characteristics of ECG and the types of arrhythmias,and then analyzes the waveform characteristics of four common types of arrhythmias.Finally,the ECG signal preprocessing,feature waveform detection and feature extraction are described in detail.The third chapter discusses the problem that the training set samples are incorrectly labeled in the automatic classification of ECG signals.Aiming at this problem,this paper makes an indepth research on the previous work,and puts forward a method based on iterative and ensemble learning based on the combination of various classification algorithms based on cross validation.The data in MIT-BIH arrhythmia database are used for the experiment.After manually modifying and labeling,the noise is added in the training set.Under different noise rates,after the iterative process combined with ensemble learning to remove the noise,the classification accuracy of the classifier is significantly improved.In Chapter 4,several classification algorithms are compared comprehensively,and support vector machine is selected to classify normal ECG signals and four common arrhythmia ECG signals.Aiming at the optimal selection of kernel parameters and penalty coefficients of radial basis kernel function in support vector machine.An improved particle swarm optimization algorithm is proposed,which not only finds the optimal parameters,but also avoids the particle swarm optimization algorithm from falling into local optimization.The experiment of ECG data in MIT-BIH arrhythmia database shows that this algorithm can classify ECG signals with high accuracy,has certain practical significance for arrhythmia analysis,and can be used for arrhythmia auxiliary diagnosis.