The invention discloses a malicious mobile application determination method based on machine learning. Whether a networking communication behavior of an application is a malicious behavior or not is automatically learnt and determined, so as to determine whether the application is malicious or not; the method relates to the field of mobile application detection and the like, and can be used for developing applications with a similar detection function; the applications can be independently installed on smartphones to use and can also support special malicious application determination detection kit and the like, which are developed by third-party detection mechanisms.