Arrhythmia (ARR) and Congestive Heart Failure (CHF) are the most common conditions that have delayed diagnoses in cardiovascular illnesses and the primary cause of death, these are compared with Normal Sinus Rhythm (NSR). Manually interpreting electrocardiogram (ECG) readings can lead to an early identification of various heart diseases. However, because ECG signals have so many different features, manual diagnosis is difficult. Patient lives could be saved with an accurate ARR and CHF group system. The signal classification problem is made simpler by the process of condensing the original signal from an ECG to a much fewer number of characteristics that work together to distinguish between several classes. The variations in variance for each of the three groups in the second-largest scale (second-lowest frequency) wavelet sub-band is examined. It makes use of a quadratic kernel multi-class SVM. This paper deals with two analyses. The whole set of data i.e. training and testing sets to determine the rate of misclassification and confusion matrix. With the best classification accuracy of 97.95%, the SVM divided the raw ECG signal data into three categories: NSR, ARR and CHF. The confusion matrix reveals the misclassification of one class to another i.e. one CHF record as NSR.