Network activities recognition has always been a significant component of intrusion detection. However, with the increasing network traffic flow and complexity of network behavior, it is becoming more and more difficult to identify the specific behavior quickly and accurately by user network monitoring software. It also requires the system security staff to pay close attention to the latest intrusion monitoring technology and methods. All of these greatly increase the difficulty and complexity of intrusion detection tasks. The application of machine learning methods based on supervised classification technology would help to liberate the network security staff from the heavy and boring tasks. A finetuned model would accurately recognize user behavior, which could provide persistent monitoring with a relative high accuracy and good adaptability. Finally, the results of network activities recognition by J48 and Na\"ive Bayes algorithms are introduced and evaluated.