A PCA-Based Change Detection Framework for Multidimensional Data Streams

Abstract

Detecting changes in multidimensional data streams is an important and challenging task. In unsupervised change detection, changes are usually detected by comparing the distribution in a current (test) window with a reference window. It is thus essential to design divergence metrics and density estimators for comparing the data distributions, which are mostly done for univariate data. Detecting changes in multidimensional data streams brings difficulties to the density estimation and comparisons. In this paper, we propose a framework for detecting changes in multidimensional data streams based on principal component analysis, which is used for projecting data into a lower dimensional space, thus facilitating density estimation and change-score calculations. The proposed framework also has advantages over existing approaches by reducing computational costs with an efficient density estimator, promoting the change-score calculation by introducing effective divergence metrics, and by minimizing the efforts required from users on the threshold parameter setting by using the Page-Hinkley test. The evaluation results on synthetic and real data show that our framework outperforms two baseline methods in terms of both detection accuracy and computational costs.

On the reliable detection of concept drift from streaming unlabeled data

“…Novelty detection/ clustering methods OLINDDA (Spinosa et al, 2007), MINAS (Faria et al, 2013), Woo (Ryu et al, 2012), DETECTNOD (Hayat and Hashemi, 2010), ECSMiner (Masud et al, 2011), GC3 (Sethi et al, 2016b) Multivariate distribution monitoring CoC (Lee and Magoules, 2012), HDDDM (Ditzler and Polikar, 2011), PCA-detect (Kuncheva and Faithfull, 2014;Qahtan et al, 2015) Model dependent monitoring A-distance (Dredze et al, 2010), CDBD (Lindstrom et al, 2013), Margin (Dries and Rückert, 2009) when it matters. Implicit drift detectors rely on properties of the unlabeled data's feature values, to signal deviations.…”

“…This has motivated the development of unlabeled drift detection techniques (Ditzler and Polikar, 2011;da Costa et al, 2016), which monitor changes to the feature distribution, as an early indicator of drift. However, existing methods using unlabeled data are essentially change detection techniques that detect any change to the data distribution, irrespective of its effects on the classification process (Lee and Magoules, 2012;Ditzler and Polikar, 2011;Kuncheva and Faithfull, 2014;Qahtan et al, 2015;da Costa et al, 2016). For the task of classification, change is relevant only when it causes model performance to degrade.…”

Optimizing Cost of Continuous Overlapping Queries over Data Streams by Filter Adaption

“…Intuitively, if the selectivity distribution is unchanged, the loading vectors 3 of the principal components are relatively stable, otherwise there can be certain rotation to the loading vectors [30]. By this means, we are able to detect the stream variation by monitoring the principal components of evaluation order vectors.…”

“…so different value range will affect the filters' significance in query execution. In this study, we tune the popularity range as [10,20], [20,30], [30,40] and [40, 50], i.e., 10% to 30% sharing degree. For each range, we generate the problem instances, create corresponding data streams, and run all strategies on the streams with simulated changing selectivity.…”

“…However, FilAda approach cannot effectively improve the performance over "NA", because the traffic status of different locations behave differently, and the stream variation will take place asynchronously, which will increase the frequency of selectivity updating for FilAda. When the popularity range increases from [0, 10] to [30,40], the average cost will natually drop because each filter can potentially resolve more queries with larger sharing degree.…”