i need paraphrasing to this text

K-Means algorithm is used to cluster the training information , and the seven clusters can be

divided into a huge cluster, five small clusters and an extremely small cluster. With cluster labels as abscissa, the average of each feature of each cluster is ordinate, they draw Fig. 2 that shows the characteristic differences between different clusters and the clustering results have strong interpretability.

So in conclusion, This paper uses CDR data to train K-Means model, then the potential category information is fond out and some feature patterns are found through the feature analysis of different clusters. Compared with other strategies the strategy in this article does not require class annotation, that decrease the difficulty of information acquisition. As the amount of information increases, the accuracy of fraud recognition increases [1].