Self-Organizing Piecewise Aggregate Approximation algortihm for intelligent detection and diagnosis of heart conditions

Gallimore, Michael, Riley, Mike and Bingham, Chris (2015) Self-Organizing Piecewise Aggregate Approximation algortihm for intelligent detection and diagnosis of heart conditions. In: International Conference on Medical and Health Sciences, 13th June 2015, Berlin, Germany.

ICMHS 2015 Proceedings Version.pdf
ICMHS 2015 Proceedings Version.pdf - Whole Document

Item Type:Conference or Workshop contribution (Paper)
Item Status:Live Archive


Electrocardiogram (ECG) signal classification is a recognized method for automated detection and diagnosis of heart abnormalities. This is typically achieved through dimensionality reduction techniques and feature extraction followed by signal classification using various machine learning algorithms. Although some algorithms can yield accurate results, they can be computationally demanding meaning that mobile analysis is difficult. Furthermore, discrete changes in signal characteristics, often exhibited as an early indication of the onset of heart abnormalities, can be lost in the dimensionality reduction process leading to misclassification of signal types. This paper presents a new dimensionality reduction algorithm, based on Piecewise Aggregate Approximation (PAA), called Self-Organizing Piecewise Aggregate Approximation (SOPAA) that is able to determine optimum PAA parameters based on signal characteristics within individual ECG data sets. This leads to more accurate and compact representations of ECG signals, improved classification of signal types and improved abnormality detection and diagnosis. In this work, ECG data from 99 patients exhibiting 3 different heart conditions are analyzed. Signals are discretized using both PAA and SOPAA and classified using the k-means clustering algorithm. It is shown that the SOPAA algorithm outperforms standard PAA by correctly classifying 19.7% more patients.

Keywords:Piecewise Aggregate ApproximationPiecewise Aggregate Approximation, ECG classification, Dimensionality reduction, optimization, bmjholiday
Subjects:G Mathematical and Computer Sciences > G760 Machine Learning
H Engineering > H900 Others in Engineering
Divisions:College of Science > School of Engineering
ID Code:17747
Deposited On:24 Jun 2015 18:53

Repository Staff Only: item control page