Tree Induction over Perennial Objects

We study the tree induction over a stream of perennial objects. The perennial objects are dynamic in nature and cannot be forgotten. The objects come from a multi-table stream, e.g., streams of Customer and Transaction. As the Transactions arrive, the perennial Customers' profiles grow and accumulate over time. To perform tree induction, we propose a tree induction algorithm that can handle perennial objects. The algorithm also encompasses a method that identifies and adapts to the concept drift in the stream. We have also incorporated a conventional classifier (kNN) at the leaves to further improve the classification accuracy of our algorithm. We have evaluated our method on a synthetic dataset and the PKDD Challenge 1999 dataset.