The Electrocardiogram (ECG) signal is by far the most intensive tool used to inspect the condition of the Heart and to detect early arrhythmia abnormalities, which is a life-saving process. The classification process highly depends on the quality of the ECG signal. Through this paper, we present a comparative study of two preprocessing techniques, namely high-pass derivative and robust neural net-work preprocessing filters. Our work involves de-veloping a Super Vector Machine (SVM) detector and assessing its performance by two preprocessing methods. We evaluated the detector's performance by using the MIT-BIH database under the AAMI EC57 standard and using Synthetic Minority Over-sampling Technique (SMOTE). The robust-based classifier shows higher performance with an overall accuracy of 99,51 % for intra-patient detection and 82,23% for inter-patient classification compared to the derivative-based one. that has an overall accuracy of 99,34% for intra-patient and 73,51 % for inter-patient detection.