TY - JOUR
T1 - Arrhythmia classification using Mahalanobis distance based improved Fuzzy C-Means clustering for mobile health monitoring systems
AU - Haldar, Nur Al Hasan
AU - Khan, Farrukh Aslam
AU - Ali, Aftab
AU - Abbas, Haider
PY - 2017/1/12
Y1 - 2017/1/12
N2 - In this paper, an improved electrocardiogram (ECG) beats classification system is proposed, which is based on Fuzzy C-Means (FCM) clustering algorithm. The classification of ECG beats is necessary in order to diagnose the type of arrhythmia (e.g., Atrial Premature Contraction (APC), Premature Ventricular Contraction (PVC), Right Bundle Branch Block (RBBB) etc.) present in the ECG records. The efficiency of any classification model highly depends on the “most relevant” set of features used. The primary goal of this study is to classify different arrhythmic beats with reduced set of relevant-only ECG attributes. The attribute selection model is based on Mahalanobis-Taguchi System (MTS); a multi-dimensional pattern recognition tool, which can dynamically choose the important set of ECG features. The number of most relevant features can vary from person to person according to the type of arrhythmia present in the respective ECG signals. A traditional Euclidian Distance (ED) based FCM can detect the spherical clusters but it may lead to improper clustering in some cases. As a solution to this problem, Mahalanobis Distance (MD) is used in the proposed model in order to improve the distance measurement procedure. In our proposed system, MD based improved Fuzzy C-Means (FCM-M) clustering is used to classify the arrhythmic beats. Experimental results show that the performance of FCM-M is significantly better than the conventional FCM for arrhythmia classification. Another direction of our proposed research is to use the concept of initial cluster centroid in order to reduce the number of program iterations. In our experiments, the number of program iterations is reduced to an average of 53% when initial centroid is assigned to FCM-M with the same classification results.
AB - In this paper, an improved electrocardiogram (ECG) beats classification system is proposed, which is based on Fuzzy C-Means (FCM) clustering algorithm. The classification of ECG beats is necessary in order to diagnose the type of arrhythmia (e.g., Atrial Premature Contraction (APC), Premature Ventricular Contraction (PVC), Right Bundle Branch Block (RBBB) etc.) present in the ECG records. The efficiency of any classification model highly depends on the “most relevant” set of features used. The primary goal of this study is to classify different arrhythmic beats with reduced set of relevant-only ECG attributes. The attribute selection model is based on Mahalanobis-Taguchi System (MTS); a multi-dimensional pattern recognition tool, which can dynamically choose the important set of ECG features. The number of most relevant features can vary from person to person according to the type of arrhythmia present in the respective ECG signals. A traditional Euclidian Distance (ED) based FCM can detect the spherical clusters but it may lead to improper clustering in some cases. As a solution to this problem, Mahalanobis Distance (MD) is used in the proposed model in order to improve the distance measurement procedure. In our proposed system, MD based improved Fuzzy C-Means (FCM-M) clustering is used to classify the arrhythmic beats. Experimental results show that the performance of FCM-M is significantly better than the conventional FCM for arrhythmia classification. Another direction of our proposed research is to use the concept of initial cluster centroid in order to reduce the number of program iterations. In our experiments, the number of program iterations is reduced to an average of 53% when initial centroid is assigned to FCM-M with the same classification results.
KW - Body area network
KW - E-Health, Arrhythmia
KW - Fuzzy C-Means clustering
KW - Mahalanobis-Taguchi System (MTS)
UR - http://www.scopus.com/inward/record.url?scp=84996937977&partnerID=8YFLogxK
U2 - 10.1016/j.neucom.2016.08.042
DO - 10.1016/j.neucom.2016.08.042
M3 - Article
SN - 0925-2312
VL - 220
SP - 221
EP - 235
JO - Neurocomputing
JF - Neurocomputing
ER -