Research On ECG Anomaly Recognition Based On Ensemble Learning Of Unbalanced Data

Cardiovascular disease is the leading cause of death and illness.Electrocardiogram(ECG)analysis serves as the most simple and effective clinical examination method for various cardiovascular diseases.At present,automatic detection is mainly achieved through training classification models.There are many abnormal labels in ECG data and the sample distributions of different labels are unbalanced,which has resulted in poor classifier performance.Simultaneously,single model-based methods often show good classification effects on partial anomalies but cannot guarantee their classification effect on all anomalies.To address this problem,a heart beat signal generation method based on generative adversarial network with attention mechanism and an ECG classification method based on feature fusion and stacking integration are proposed.They address the unbalanced ECG data and the poor generalization ability of single model,respectively.The main research contents are as follows.(1)A heart beat signal generation method based on generative adversarial network with attention mechanism was proposed to address the problem of unbalanced distribution of ECG data samples.In this method,the heartbeats of all samples are segmented.And then,a generative model and a discriminative model are constructed to form a generative adversarial network,and an attention mechanism is introduced into the discriminative model for heartbeat signal generation.Finally,the ECG data enhancement method based on generative adversarial network is compared with other enhancement methods.The results show that the data enhancement method based on generative adversarial network has better classification performance and can effectively solve the problem of the unbalanced sample distribution in ECG data.(2)In view of the fact that a single model cannot classify all the abnormalities with desired accuracy,an ECG classification method based on feature fusion and stacking integration is proposed.On one hand,some features are extracted based on human’s prior knowledge.On the other hand,the features hidden in the convolutional neural network are extracted by machine learning.The extracted features with different resolutiuons are then fused and processed further.Finally,stacking integration was used to classify the fused ECG features,and the experimental results show that the classification accuracy can be effectively improved by the optimized selection and ensemble of different classifiers through stacking.The main contribution of this research is as follows.On one hand,the problem of unbalanced sample distribution of ECG data was solved effectively by the generation of ECG data through generative adversarial network.Meanwhile,attention mechanism was put forward in the generative adversarial network to extract the significant region with abnormal target,which maks the generated samples appromixate the real distribution precisely.On the other hand,ECG features extracted with different scales are fused and the ensemble learning was exploited for classification based on the fused features,thus improving the classification accuracy and generalization performance...