The Research on Tracking Concept Drift Based on Genetic Algorithm

摘要

Concept drift is a crucial problem in machine learning. The machine learning model, such as classifier, should adapt to the concept change rapidly to keep the precision. At present, most methods for solving this problem is based on the mechanism, which is referred to as Time -Weight. However, current existing methods can not quickly adapt to the concepts, which reappear. Therefore, a method with the ability to adapt concept changing all the time is necessary, especially for the situation of time precious, such as data stream classification. In this paper, we proposed a method of tracking concept drift based on genetic algorithm (TCDGA). We experimented with TCDGA on public dataset and made comparisons. Our results show that TCDGA can yield better performance when concepts reappear.