A Comparison of Statistical Machine Learning Methods in Heartbeat Detection and Classification

Abstract

PrefaceThe arrival of the so-called Petabyte Age has compelled the analytics community to pay serious attention to development of scalable algorithms for intelligent data analysis. In June 2008, Wired magazine featured a special section on "The Petabyte Age" and stated that "..our ability to capture, warehouse, and understand massive amounts of data is changing science, medicine, business, and technology." The recent explosion in social computing has added to the vastly growing amounts of data from which insights can be mined. The term "Big Data" is now emerging as a catch-all phrase to denote the vast amounts of data at a scale that requires a rethink of conventional notions of data management.There is a saying among data researchers that "more data beats better algorithms." Big Data provide ample opportunities to discern hitherto inconceivable insights from data sets. This, however, comes with significant challenges in terms of both computational and storage expense, of the type never addressed before. Volume, velocity, and variability in Big Data repositories necessitate advancing analytics beyond operational reporting and dashboards. Early attempts to address the issue of scalability were handled by development of incremental data mining algorithms. Other traditional approaches to solve scalability problems included sampling, processing data in batches, and development of parallel algorithms. However, it did not take long to realize that all of these approaches, except perhaps parallelization, have limited utility.The International Conference on Big Data Analytics (BDA 2012) was conceived against this backdrop, and is envisaged to provide a platform to expose researchers and practitioners to ground-breaking opportunities that arise during analysis and processing of massive volumes of distributed data stored across clusters of networked computers. The conference attracted a total of 42 papers, of which 37 were research track submissions. From these, five regular papers and five short papers were selected, leading to an acceptance rate of 27%.Four tutorials were also selected and two tutorials were included in the proceedings. The first tutorial entitled "Scalable Analytics: Algorithms and Systems" addresses implementation of three popular machine learning algorithms in a Map-Reduce environment. The second tutorial, "Big-Data: Theoretical, Engineering and Analytics Perspectives," gives a bird's eye view of the Big Data landscape, including technology, funding, and the emerging focus areas. It also deliberates on the analytical and theoretical perspectives of the ecosystem.The accepted research papers address several aspects of data analytics. These papers have been logically grouped into three broad sections: Data Analytics Applications, Knowledge Discovery Through Information Extraction, and Data Models in Analytics. VI PrefaceIn the first section, Basil et al. compare several statistical machine learning techniques over electro-cardiogram (ECG) datasets. Based on this study, they make recommendations on feat...

Leveraging a discriminative dictionary learning algorithm for single-lead ECG classification

“…Adhering to common practice in heartbeat classification, we used performance indices metrics for accuracy (Acc), sensitivity (Se), positive predictive value (PPV) and false positive ratio (FPR). It is observed in Table 1 that our algorithm gives competent results for sampling rate of 360 Hz in comparison to state-of-the-art results obtained in [2], while also providing discriminatory results at 114 Hz as compared to those of [7], thereby achieving the dual goal of compression and accuracy.…”

Feature Extraction and Automated Classification of Heartbeats by Machine Learning

“…Alvarado et al [13] in a departure from traditional approaches used pulse based representations of signals using time based samplers such as Integrate and Fire (IF) model [13]. In [14], we compared the performance of LDA, QDA and artificial neural networks (ANN) in detecting Ventricular ectopic Beats (VEB). In [15], we focused on detecting Supraventricular ectopic Beats (SVEB) and proposed a classification technique based on the variations in the ECG morphology of SVEBs.…”