Find the right transaction length for stream mining: A distance approach

Stream data mining has drawn people´s attention for the last decade. Different algorithms have been proposed and applied in different areas. Most of the stream data mining algorithms are use a sliding window to cache the stream during mining. Most research have been focused on statically or dynamically generate the sliding window, yet the proper selection of the transaction length have not been addressed. Transaction length decides the length the pattern found in a stream and affect the mining processing time as well. This paper proposed a distance method to evaluate the proper transaction length value in mining process. Experiment demonstrated that this method could successfully find the pattern length in emulated telecommunication stream data. By using this method in data pre-processing, it could find a suitable transaction length value for the mining process which could make mining more efficient therefore improve the performance.