Incremental-Decremental Attribute Learning Algorithm Based on K-prototypes for Mixed Data Stream Clustering

Abstract

In the view of the high requirement of memory space, the long execution time and the low execution efficiency of the k-prototypes algorithm in large-scale training samples, this paper puts forward a new online incremental and decremental learning algorithm based on k-prototypes in order to deal with mixed data sets. For deep understanding, this study provides a better insight into one particular machine learning algorithm the incremental attribute learning based on unsupervised clustering algorithm and tackles also the decremental attribute learning task while being crucial and essential in data stream context. Indeed, this entering mixed data is either escorted with new attributes, or incorporates less attributes than the old ones. So, it might be processed sequentially over flexible time windows, in less time consuming and with providing a well-defined model with better separation between clusters. The efficiency of our proposed online learning algorithm is proved across the experimental results based on different evaluation criteria and numerous simulations made on real mixed data sets. Keywords: Clustering, data stream, mixed attributes, k-prototypes, incremental-decremental attribute learning, merge.