Abstract: Mining concept drifts from data streams is one of the most important fields in data mining. In this paper, a new mine algorithm called M_ID4 is proposed, which aims at quickly detecting drifted concepts from a large volume of data stream by using a small training data set. M_ID4 uses ensemble multi-classifiers to mine concept changes from the data streams, and its every classifier in the ensemble is an improved ID4 algorithm with an incremental way. The experimental results show that M_ID4 algorithm is of higher accuracy and better adaptability to quick drifted concepts than the popular algorithms.