Best Parameters Selection of Arrhythmia Classification Using Convolutional Neural Networks

Abstract

Arrhythmia are disturbances in the heart where the heart beats slower or faster. Some types of Arrhythmia can became a serious problem and life-threatening. Early detection of Arrhythmia is very crucial to patients. Tools that can be used to determine heart condition is Electrocardiogram (ECG). Deep learning methods can be used to classify types of Arrhythmia from ECG images. Convolutional Neural Network is one of deep learning methods that is often used to classify images. CNN-based model such as VGG, ResNet, and MobileNet has gotten success in images classification. Those models are using lots of convolution layer, so those models are easily run into over fitting problem if those are used in small dataset. CNN model in this research needs parameter adjustments to solve over fitting problem. Parameter that were being adjusted were learning rate, dropout rate, and the number of convolution layer. The testing results on CNN model showed that the best learning rate and dropout rate which produced the best model to classify Arrhythmia were 0.0001, and 0.0075 respectively. The number of convolution layers which obtained the best accuracy was 4. Classification using CNN model for Arrhythmia with learning rate, dropout rate, and number of convolution layers were 0.0001, 0.0075, and 4 respectively resulted in the best model with 94.2 % accuracy value.