The invention relates to a typical behavior real-time identification method based on a wireless wearable type identification and sensing platform, and belongs to the Internet of Things intelligent household field. Acceleration, time, radio frequency angles, channels, receiving signal intensity characteristics are extracted to be used as input of a behavior prediction phase. In the behavior prediction phase, by considering a class imbalance problem, a dynamic weighted conditional random field is used to introduce a dynamic calculation class-related weight parameter to a target function in order to punish misclassified minority classes. An evaluation function is provided in a behavior identification phase of leaving a bed or a chair, and the target thereof is to distribute an activity class to data occupying a leading position in one second, and errors caused by noises in the prediction phase are reduced. When the behavior of the user leaving the bed or the chair is identified, alarms are given, and at the same time, the alarms of the same class are not given in the next 1.75seconds. Tumbling of elders is effectively prevented, and real-time feedback information is provided for the nursing of the elders.