Paroxysmal Atrial Fibrillation (PAF) is a very common rhythm disorder that causes rapid and irregular impulses in the heart. In this study, it is aimed to determine whether patients can be warned before PAF events. 30-minute HRV data used in this study. Each piece of data was divided into 10 pieces of 5-minute parts. Time domain measurements from linear measurements of HRV and Poincare measurements from nonlinear measurements of HRV were used for each segment. Detecting performances were measured for each segment using k-nearest neighbor classifier. Particularly linear measurements have been shown to achieve up to 82% success in predicting PAF attack and was observed that PAF attack could be detected 12,5 minutes earlier.