To address the inaccuracy and high time complexity of traditional data stream mining technology, this paper introduces a new algorithm of date detection based on k-distance to pruning and comentropy to detect sliding windows. When the data fills the current window, the k-distance of the data is used to prune all data in the pruning time. As a result, most normal data is filtered out. Experimental results demonstrate that the SWKC algorithm possesses better efficiency and accuracy than some other traditional detection algorithms.