Early Abnormal Heartbeat Multistage Classification by using Decision Tree and K-Nearest Neighbor

Heart diseases contribute to the highest cause of death around the world particularly for middle aged and elderly people. There are various types of heart disease symptoms. One of the most common types is Arrhythmia which is considered as a dangerous heart condition since the symptom itself may initiate more chronic heart diseases and result in death if it is not treated earlier. However, the detection of Arrhythmia by humans is regarded as a challenging task because the natures of the symptom appear at random times. Therefore, an automatic detection method of abnormal heartbeat in ECG (electrocardiogram) data is needed to overcome the issue. In this paper, a novel multistage classification approach using K-Nearest Neighbor and decision tree of the 3 segments in the ECG cycle is proposed to detect Arrhythmia heartbeat from the early minute of ECG data. Specific attributes based on feature extraction in each heartbeat are used to classify the Normal Sinus Rhythm and Arrhythmia. The experimental result shows that the proposed multistage classification approach is able to detect the Arrhythmia heartbeat with 90.6% accuracy for the P and the Q peak segments, 91.1% accuracy for the Q, R and S peak segments and lastly, 97.7% accuracy for the S and the T peak segments, outperforming the other data mining techniques.