Chapter three - Generalization performance of deep autoencoder kernels for identification of abnormalities on electrocardiograms

Abstract

Listen

Translate article

Modeling of deep learning (DL) structures with many hidden layers and a large number of neurons at each layer for the data analysis process has brought about the requirements for long training time and computation capacity depending on the increase in the number of optimization parameters. This chapter addresses the problem of how to reduce the training time required for DL algorithms by combining theories in deep autoencoder kernels. Existing studies in DL tend to focus on analyzing images to obtain visual arrangement and feature learning. The DL algorithms are used to handle advanced-level image processing including image generation, pose recognition, image captioning, character recognition, and more; and these need a long training process. Most research in the literature covers the application of DL on different patterns. However, little research has addressed the issue of simplifying the mathematical background while maintaining the generalization capacity and robustness of the classifiers. As a result, the aim of this chapter is to provide an overview of how the training time and generalization capability for DL algorithms including deep belief networks (DBNs) and extreme learning machine (ELM) autoencoder kernels. The DL algorithms are used to diagnose coronary artery disease (CAD) on the electrocardiogram (ECG) and fiducial features. The proposed DBN model achieved classification performance rates of 96.93%, 96.03%, and 91.23% for accuracy, sensitivity, and specificity. The best contribution is achieved using an optimized ELM autoencoder among DL algorithms.