A typical method for capturing electrical activity of human heart is electrocardiography (ECG). Because of its low amplitude and extreme susceptibility to high frequency noise and other distortions, accurate computer analysis of ECG signals is difficult. This paper's main goal is to provide machine learning methods for diagnosing myocardial infarction and differentiating between arrhythmias, hypertrophy, and heart enlargement. This research propose novel technique in ECG signal processing based on machine learning architectures by feature extraction with classification. Here the input ECG signal is processed for noise removal, smoothening and normalization; signals has been obtained as frames. Then this signal feature has been selected using support vector machine integrated with backward elimination architecture. the selected features of ECG signal is classified using probabilistic particle swarm optimization. From the classified ECG signal the type of cardio vascular disease detected. The experimental analysis is carried out for different dataset in which the proposed technique is analysed in terms of accuracy, precision, recall, F-measure, RMSE.