Classification of cardiac arrhythmias using ID3 classifier

Dissertant

Shukr, Nidal Hamid

Thesis advisor

Sadiq, Ahmad T.

University

University of Technology

Faculty

-

Department

Department of Computer Engineering

University Country

Iraq

Degree

Master

Degree Date

2013

English Abstract

Electrocardiogram (ECG) is the record of the heart muscle electric impulses.

Received and processed ECG signal could be analyzed, and results could be used for the detection and diagnostics of cardiovascular diseases (CVD).

One of the important CVD diseases is arrhythmia.

Accurate detection of ECG features is an important demand for medical purposes, therefore an accurate algorithm is used to detect these features.

In this thesis an approach has been proposed to detect classify the cardiac arrhythmia from a normal ECG signal based on wavelet decomposition and Iterative Dichotomiser3(ID3) algorithm.

The first stage takes several ECG signals, these signals are preprocessed by using a zero-padding which is a simple scheme based on signal extension on the boundaries of the signal.

In the second stage, vectors of five features are extracted by the computation of one dimensional wavelet transform of all input ECG signals once using Haar filter and again using Daubechies4 (Db4) filter, these features involve time durations and amplitudes of selected ECG signals.

In the third stage the numerical values of feature vectors that obtained from the second stage are represented in the form of nominal values within a data nominalization sub stage and then using this data as training input dataset for ID3 algorithm to learn and produce a decision tree (classifier).

Then this classifier is converted to a set of rules which is using to classify five types of ECG arrhythmias including normal case.

The Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia Database is utilized to learn and test the classifier.

The experimental results show that the ID3 classifier achieves accuracy of 92% in the case of Haar transform and 94% with Db4 transform.

Main Subjects

Medicine

Topics

• Heart
• Diseases
• Electrical properties
• Diagnosis
• Electrocardiography

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-418265