Detecting concept change in dynamic data streams

Abstract

In this research we present a novel approach to the concept change detection problem. Change detection is a fundamental issue with data stream mining as classification models generated need to be updated when significant changes in the underlying data distribution occur. A number of change detection approaches have been proposed but they all suffer from limitations with respect to one or more key performance factors such as high computational complexity, poor sensitivity to gradual change, or the opposite problem of high false positive rate. Our approach uses reservoir sampling to build a sequential change detection model that offers statistically sound guarantees on false positive and false negative rates but has much smaller computational complexity than the ADWIN concept drift detector. Extensive experimentation on a wide variety of datasets reveals that the scheme also has a smaller false detection rate while maintaining a competitive true detection rate to ADWIN.

Capturing recurring concepts using discrete Fourier transform

“…False positive rate is controlled by adjusting the user parameter significance level of a hypothesis test. Because of multiple passes on data in the current histogram, false positive rate, especially in a noisy environment, is relatively higher than SeqDrift2 .…”

“…In contrast, SeqDrift2 was stable at these points in the stream. In , it was shown that ADWIN's false positive rate was higher than that of SeqDrift2 and false positives signaled by ADWIN can cause existing spectra that are performing well to be switched to a new spectrum which under performs. This problem was particularly acute at low‐memory setting of pool Size 1.…”

“…All other parameters are automatically optimized minimizing detection delay. SeqDrift2 is empirically shown to perform better in terms of false positive rate, processing time and noise tolerance in [7]. Now, we will describe the DFT application process with a toy example.…”

“…Both FCT algorithms were almost equivalent on 20% noisy dataset. SeqDrift2 as shown in [7] is more resilient to noise than ADWIN. This property of SeqDrift2 helped FCT to achieve the best accuracy on 30% noise level.…”

“…In contrast, SeqDrift2 was stable at these points in the stream. In [7], it was shown that ADWIN's false positive rate was higher than that of SeqDrift2 and false positives signaled by ADWIN can cause existing spectra that are performing well to be switched to a new spectrum 4032 S. SRIPIRAKAS AND P. RUSSEL FCT, Fourier concept trees. As all algorithms have a forest of trees and a Fourier pool, memory consumption is divided into two columns for each pool size experimented.…”