This study identified appropriate predictive methods for numbers of video views of online courses. Using small samples, we analyze and compare the predictive performance of the three methods, including grey prediction least squares method, GM(1,1), and support vector regression . We employed the actual views of 43 online courses (from July 2017 to June 2018) as training models and verified the number of visitors using data in July, 2018 from a national university in Taiwan. The results of the three methods demonstrated consistent predictive power for the top 15 courses, whereas 3 courses were unpredictable. In addition, for the last 15 courses, the GM(1,1) model accurately predicted the actual number of views observed, whereas the least squares method and support vector regression generated errors. This study provides a valuable reference for educational institutes to maximize predictive accuracy by increasing the number of visitors.