Abstarct: Annually, many people die by cardiac murderous arrhythmias. Most of deathly arrhythmias happens in ventricle. Ventricular tachycardia (VT) and ventricular fibrillation (VF) are the most common symptoms causing to sudden cardiac death (SCD) especially in the patient with history of heart diseases. Many studies have been carry out to predict VT-VF, so to prevent SCD. Implantable cardioverter defibrillators (ICDs) is one of the recent devolved devices used to send electrical shock to prevent SCD. ICDs devices implanted in the chest and always accompany the patient. In this thesis, two novel method baxsed on output signal of ICDs (called Heart Rate Variability (HRV)) are proposed to predict VT and VF arrhythmias. Considering the amount of noise in HRV signals, ectopic beats were firstly removed using wavelet transformation and median filter. Extracted features makes two proposed methods. In first method, considered features such as short-time autocorrelation, short-time signal energy and short-time zero crossing have then been extracted to use for prediction with support vector machine (SVM). The results obtained indicate 82.6% accuracy through the constructed model. This results has little improvement in comparison with previous works. Second method features baxsed on Fourier-Bessel (FB) expansion had a good history in analyzing non-stationary signals. By using the FB coefficients, we improve the result to 93% in accuracy. This reveals significant improvement and efficiency in detecting VT-VF compared to what has been done before. This performance, severely can effected on reducing inappropriate shocks and make a better output for appropriated treatment.