Double Window Concept Drift Detection Method Based on Sample Distribution Statistical Test

摘要

Aiming at the problem of data stream drift due to its time attribute, this paper proposes a double window concept drift detection method based on sample distribution statistical test. In this method, support vector regression (SVR) is used to detect data anomalies in the outlier detection window. The Euclidean distance between samples is statistically tested in the distribution detection window to reflect the change of data distribution. This method essentially explores the concept drift phenomenon and applies it to the regression problem. This paper verifies the performance of this method based on the benchmark data set of concrete compressive strength.