Abstract:According to the lack of efficient analysis tool for Quick Access Recorder(QAR) data, an improved data mining method is proposed in this paper. First, a modified algorithm of k-means based on probability theory is given. Then the cluster number of QAR data set is determined, so that better cluster results can be obtained. In order to identify the atypical data and the class of typical data, a weighted minimum distance classification as well as probability analysis is used. At last, experiments of cluster and classification are given to indicate the feasibility and effectiveness of the new method.