Research On KNN And SVM Applying To ECG Classification

Because of the suitability for small samples and the well generalization performance, SVM has been become the new international research hotspot of pattern recognition field at present. The object of the paper is on the one hand to analyze ulterior SVM classified principle and on the other hand propose new algorithm for some limitation in application. An improved algorithm that applying SVM to ambiguities was proposed, which developed SVM application scope. The contributions of this dissertation are as follows:(1) An new method to deal with the ECG dataAccording to the characteristic of the ECG data. A new method to use the samples is presented: training each of the eight conductors belonging to a sample in the whole set one by one. Then give a comprehensive judgment based on that of the eight conductors. We make a contrast between this method and another two methods t lat connect eight conductors continuously or only use one conductor. Experiments show that by the eight conductors parallel training method, not only the accuracy is improved, but also the perf jrmance of this algorithm is of high stability. The reason is because it highly decrease the sample s dimension in the feature space from 344 to 43, and consider the eight conductors comprehensively as well so as to decentralize the risk and mistaken factors.(2) KNN -SVM classified algorithmSVM focus on the samples near the boundaiy in training time, and those samples intermixed in another class are usually no good to improve the classifier's performance, instead they may greatly increase the burden of computation and their existence may lead to ovefitting and decrease the generalization ability. In order to improve the generalization we present an improved SVM: KNN-SVM. It first prunes the training set, reserve or delete a sample according to whether its K nearest neighbor has the same class label with its elf or not, then train the new set with SVM to obtain a classifier. Experiment results show that KNN-SVM is better than SVM in speed and accuracy of classification.