Electrocardiogram (ECG) is a non-stationary signal which constitutes a lineal recording that provides an insight into heart's electrical activity. Because Cardio Vascular diseases (CVD) are numbered one as reason of mortality globally, detection of abnormalities in ECG at an early stage is crucial for diagnosis and accordingly the treatment. A system can help the cardiologists in diagnosing the arrhythmia present in a patient. The proposed architecture uses Adaptive Neuro-Fuzzy Inference System (ANFIS) with the input preprocessed with subtractive clustering method to learn fuzzy logic. Five morphological and five statistical ECG features are utilized to determine if the patient's heartbeats are normal or irregular and classify them accordingly. During statistical feature extraction, Principal Component Analysis (PCA) on detail coefficients is implemented for optimization. For classification, four classes of ECG are considered, Left Bundle Branch Block (LBBB), normal, Atrial Premature Contraction (APC) and Paced Beats (PB). The proposed system gives an overall classification accuracy of 97.75%. The overall sensitivity, average specificity and average false prediction ratio obtained are 0.9775, 0.9925 and 0.0075 respectively. © 2017 IEEE.