Caucus-based transaction clustering
Jinmei Xu, Sam Yuan Sung · 2003
Transaction clustering has received attention in recent developments of data mining. Traditional clustering methods are not useful to solve this problem. Transaction data sets are different from the traditional data sets in their high dimensionality, sparsity and numerous outliers. We introduce a new efficient algorithm for transaction clustering. The proposed algorithm is based on a caucus, which is fine-partitioned demographic groups based on purchase features of customers. Due to the important role caucus plays, we also present a heuristic method of caucus generation with the use of entropy. Experiments on real and synthetic data sets show that our approach can achieve a better result than existed methods.