This study explores the variability in stay point detection accuracy influenced by GPS log intervals and detection algorithms and confirms the impact of stay point detection on stay duration and trip frequency estimations. We compare three major detection algorithms across varied log intervals adjusted through down-sampling using five evaluation indices newly proposed in this study. Using GPS trajectory data with ground truth data collected in Hiroshima, Japan, we found that ST-DBSCAN, a time-distance density clustering method, offers the highest accuracy and maintains its performance up to a 5-minute interval. We also found that widely used conventional methods, including duration and distance-based methods and DBSCAN, would produce considerably biased results.