ECG Classification using Wavelet Transform and Discriminant Analysis Farah Nur Atiqah Francis Abdullah, Fazly Salleh Abas, Rosli Besar Faculty of Engineering and Technology Multimedia University Jalan Ayer Keroh Lama 75450 Melaka Malaysia marianne_julie@yahoo.co.uk, fazly.salleh.abas@mmu.edu.my, rosli@mmu.edu.my Abstract—This paper focuses on two cardiac conditions, the supraventricular ectopy and the ventricular ectopy. Four different mother wavelets are used to produce sets of features. Results shows that each cardiac conditions beat has its own unique characteristics and also decomposition of different mother wavelet produced different degree in discriminative power. The Discriminant Analysis Classifier of different distance metric (linear, quadratic and mahalanobis) are tested. Classification performance mostly reached more than 90% for both individual feature and combined feature classification. Keywords-cardiac conditions; wavelet; discriminant analysis; distance metric I. INTRODUCTION Cardiac problem has been a major health issue in these present days considering the amount of admission caused by the cardiac disease. Many studies have been done to assist the prevention and early detection of the cardiac cases among people. Thus, a lot of new algorithms have been developed and used for that purpose. ECG beat recorded from a person proved to be very useful in determining types of cardiac disease that an individual is suffering from because each and every one of the cardiac problem may produce different patterns of ECG signal. This also proved to be useful to distinguish which cardiac problem class does a person’s heartbeat belongs to. The electrocardiogram (ECG) is a signal with unique representations of bioelectric potential with respect to time which can be called as the heartbeat. Figure 1 shows a typical ECG beat pattern according to [1]. Its unique behavior can be a very useful tool for the diagnosis of cardiac disorder. However, the classification of ECG is quite time consuming and tedious because different patient with the same cardiac problem may produce different shape and regularity in its signal. Which means the conduction of electric impulse varies from one patient to another. Detection algorithm of the ECG signal is studied in [3] using wavelet transform decomposition property. The R peak of each ECG signal is very important such as it is one of the earliest processing steps before it undergoes signal analysis. Various methods have been used to extract features from the ECG signal to provide identifications of different classes of heartbeat for signal classification. Each method produces different degree of results depending on the features used for the classification. The most important features to be included in the feature selections are the RR intervals features. Figure 1 shows the typical pattern on an ECG signal together with its P component, QRS component and T component annotations for clearer view. These features are always combined with other additional features to enhance classification performance. The classifier used for the classification of the ECG signal also plays an important role. Selected classifiers performances were tested first to determine the best classifier used that is suitable for classification using those particular features. This is because different group of features may give out different classification performance when tested with different type of classifier. Figure 1. ECG signal pattern [1] Effective and detail study for feature extractions and classification of ECG signal is necessary to produce the strong and stable classifier for ECG signal classification regardless of the condition of the signal. 2012 International Conference on Biomedical Engineering (ICoBE),27-28 February 2012,Penang 978-1-4577-1991-2/12/$26.00 ©2011 IEEE 191