Fining Frequent Itemsets from Uncertain Transaction Streams

Uncertainty pervades many application domains such as pattern recognition, sensor networks and mobile object tracking. However, in those applications, uncertain data often arrives at high speed and need to be processed in a streaming fashion. Frequent itemset mining is one of the most common problems when analyzing those uncertain transactions in streaming data. In this paper, we propose an efficient algorithm based on possible world semantics, called FI-UTS (Frequent Itemsets mining in Uncertain Transaction Streams), for finding the set of all frequent itemsets from the uncertain streaming data with a sliding window. A novel decreasing maximum count function in the algorithm is proposed to reduce the running time and the number of frequent itemset to be maintained when the window slides forward. Experimental results show that FI-UTS algorithm is much better than some methods for mining frequent itemsets in uncertain streams.