Select start point for ARF analysis
Kenichi Yoshida · 2024
It is important to analyze online data whose characteristic changes over time, such as financial and coronavirus infection data. Many studies have been conducted. In general, ensemble-based learning methods perform well as analysis methods, and methods such as SEA, DWM, and ARF have been proposed. In addition, the change in characteristics over time is called concept drift, and its classification and detection methods have been studied. This paper reports the adverse effects of a characteristic that has yet to be considered in the classification of concept drifts and proposes a solution. The characteristic discussed in this paper is the ratio of explanatory variables whose characteristics change. This paper shows that when this ratio is large, it harms the ensemble-based learning methods. In addition, this paper evaluates a solution method to show that the method can improve the accuracy of the analysis. Since a naive implementation of the proposed method is inefficient, this paper also reports a beam search implementation.