Nowadays Android smart mobile devices have become the main target of malware developers, so detecting and preventing Android malware has become an important issue of information security. Therefore, this paper proposes an Android application classification system that combines static permissions and dynamic packet analysis. This system first obtains the static information of Android applications through static analysis, classifies the applications as benign or malicious through machine learning, and avoids excessive dynamic data collection time by filtering out benign applications. Then in the dynamic analysis stage, the malware's network traffic is used to extract multiple types of features, and then machine learning is used to achieve the malware family classification. The experimental results showed that the accuracy rate of the static model for malicious and benign classification was 98.86%. On the other hand, the accuracy of the dynamic model proposed in this paper for family classification of applications is 96%, which is better than 94.33% of DroidClassifier . The final experiment confirmed that the system proposed in this paper can not only save 52.5% of dynamic data collection time but also improve the accuracy of Android application family classification.