Many data mining applications analyze structured data that span across many tables
accumulate in time. Incremental mining methods have been devised to adapt patterns to new tuples. However, they have been designed for data in one table only. We propose a method for incremental clustering on multiple interrelated streams - a “
”: its components are streams that reference each other, arrive at different speeds and have attributes of a priori unknown value ranges. Our approach encompasses solutions for the maintenance of cach-es and sliding windows over the individual streams, the propagation of foreign keys across streams, the transformation of all streams into a single-table stream, and an incremental clustering algorithm that operates over that stream. We evaluate our method on two real datasets and show that it approximates well the performance of an ideal method that possesses unlimited resources and knows the future.