With the rapid development of crowdsourcing, more and more Internet companies choose to crowdsource their software tasks. However, software tasks have their own characteristics, such as high threshold, high complexity, and long period, which make them face serious problem of fewer participants. In this paper, using the data on TopCoder, which is the world's largest crowdsourcing platform for software, we carefully researched the quantity of participants of crowdsourcing software tasks. Firstly, we analyzed the factors affecting participation of crowdsourcing software tasks by multiple regression method. Then, participation prediction model was studied with classification algorithms in data mining area. We hope that this empirical study could help companies or crowdsourcing platforms reduce the risk of low participation in crowdsourcing software tasks.