Because of the concept drift problem hidden in the uncertain data stream, it is very difficult to classify them effectively. Based on the general decision tree algorithm, the adaptive fast decision tree algorithm can count information gain based on the adaptive learning rule, and detect uncertain numerical attributes though the principle of the non-marking learning scene. The numerical attribute is transformed into a non-determined classification attribute by using splitting method, so classification of uncertain data stream is realized effectively. Then the concept drift phenomenon is effectively detected in the uncertain data stream. Simulation results show the reliability of the proposed method.