In the era of mobile Internet, a vast amount of geo-spatial data allows us to gain further insights into human activities, which is critical for Internet Services Providers (ISP) to provide better personalized services. With the pervasiveness of mobile Internet, much evidence show that human mobility has heavy impact on app usage behavior. In this paper, we propose a method based on machine learning to predict users' app usage behavior using several features of human mobility extracted from geo-spatial data in mobile Internet traces. The core idea of our method is selecting a set of mobility attributes (e.g. location, travel pattern, and mobility indicators) that have large impact on app usage behavior and inputting them into a classification model. We evaluate our method using real-world network traffic collected by our self-developed high-speed Traffic Monitoring System (TMS). Our prediction method achieves 90.3% accuracy in our experiment, which verifies the strong correlation between human mobility and app usage behavior. Our experimental results uncover a big potential of geo-spatial data extracted from mobile Internet.