Classification of Multi Heart Diseases With Android Based Monitoring System


Electrocardiogram (ECG) examination via computer techniques thatinvolve feature extraction, pre-processing and post-processing was implementeddue to its significant advantages. Extracting ECG signal standard features thatrequires high processing operation level was the main focusing point for manystudies. In this paper, up to 6 different ECG signal classes are accuratelypredicted in the absence of ECG feature extraction. The corner stone of theproposed technique in this paper is the Linear predictive coding (LPC)technique that regress and normalize the signal during the pre-processingphase. Prior to the feature extraction using Wavelet energy (WE), a directWavelet transform (DWT) is implemented that converted ECG signal tofrequency domain. In addition, the dataset was divided into two parts , one fortraining and the other for testing purposes Which have been classified in thisproposed algorithm using support vector machine (SVM). Moreover, using MITAI2 Companion was developed by MIT Center for Mobile Learning, theclassification result was shared to the patient mobile phone that can call theambulance and send the location in case of serious emergency. Finally, theconfusion matrix values are used to measure the proposed classificationperformance. For 6 different ECG classes, an accuracy ration of about 98.15%was recorded. This ratio became 100% for 3 ECG signal classes and decreasesto 97.95% by increasing ECG signal to 7 classes.