An Exploration of Graph Based Approaches in Data Stream Mining

Progressive advancements in the data stream mining have paved way for many algorithms to solve the problem of concept drift, a serious problem due to the wavering nature of the real time concepts. Concepts tend to change with time therefore concept drift is unavoidable in data stream mining but efficient algorithms can be designed to detect the concept drift and solve the problem. This paper presents a comparison based frame work of five algorithms that are designed to elucidate the drift. The five algorithms presented are OLIN (online information network), VFDT (very fast decision tree), CVFDT (concept adapting very fast decision tree), UFFT (ultra fast forest trees), and CBDT (concept based decision tree). All the above said frame works are based on sliding window algorithm and decision trees. Each time when the concept changes both the old and new set of concepts or data streams are compared to check their rigidity, the more rigid one continues to exist and the other is eradiated. To determine this disparity, decision trees are used as models. Construction of the trees and ability to grow the tree is the distinguishing factor of all the algorithms. The aim of the paper is to effectively discuss the algorithms and suggest the best out of the list, comparing their efficiencies and performances depending on the construction of the tree, its size and various other factors. Keywords-Data stream mining; Decision tree; Sliding window; Training model; Fuzzy network; Linear separability.