FEDD: Feature Extraction for Explicit Concept Drift Detection in time series

Abstract

Abstract-A time series is a sequence of observations collected over fixed sampling intervals. Several real-world dynamic processes can be modeled as a time series, such as stock price movements, exchange rates, temperatures, among others. As a special kind of data stream, a time series may present concept drift, which affects negatively time series analysis and forecasting. Explicit drift detection methods based on monitoring the time series features may provide a better understanding of how concepts evolve over time than methods based on monitoring the forecasting error of a base predictor. In this paper, we propose an online explicit drift detection method that identifies concept drifts in time series by monitoring time series features, called Feature Extraction for Explicit Concept Drift Detection (FEDD). Computational experiments showed that FEDD performed better than error-based approaches in several linear and nonlinear artificial time series with abrupt and gradual concept drifts.

Handling Concept Drifts in Regression Problems – the Error Intersection Approach

“…Existing approaches for drift detection on regression problems focus on the computation of dedicated drift detection features on the input data in order to detect drifts [19]. In contrast to this, we want to develop an approach based on the prediction error of various models in regression problems.…”

Data Mining Models of High Dimensional Data Streams, and Contemporary Concept Drift Detection Methods: a Comprehensive Review

“…The work [33] presents that the commonly selected sequence models are Hidden Markov Models which have limitations of defining simple patterns. The referred work [34] states the utilization of online algorithm to detect the change of occurrence of change at any particular time point to study the data [35], [36], [37] by Bayesian Online Change Point Detection BOCPD [34]. However, these algorithms also need a prior model.…”

“…[2], [7], [22] Surveys [27] Network-intrusion-detection field Unsupervised [23] Univariate Outlier-detection [24], [25], [26] Multivariate By example models, can't react to the changes occurred in the correlated components considered [27], [28] Multivariate solo monitoring of components using ensemble scalar CDT, every single component of the data stream is inspected to detect concept drift in a multivariate data, can't react to the changes occurred in the correlated components considered [29], [30] Multivariate nonparametric density models [31], [32], [26] Multivariate 'Pure' detectors, outperform with a low volume of data Supervised [33] Machine learning Uses commonly selected sequences as Hidden Markov Models, limits to define simple patterns. [34], [35], [36], [37] Machine learning Uses online algorithm, performs low on high dimensional data [34] Machine learning Bayesian Online Change Point Detection BOCPD, performs low on high dimensional data [41] Machine learning Extension of BOCPD, detects gradual changes [38], [39] Machine learning Not responsible for a gradual change. [40] Machine learning Detecting the gradual change points, analyzes error prediction of a learned model.…”

On-Line Dynamic Time Warping for Streaming Time Series

“…We say that a concept drift has occurred when the time series changes from one stationary interval to another. In order to adapt the DTW measure to the concept drift phenomenon, we propose a simple yet efficient weighted memory mechanism that modulates the contribution of the point-wise distance d i,j in the weight of a path (see Equation (1)). This mechanism leverages the assumption that recent points are more likely to have been produced in the last stationary interval.…”

“…These stringent conditions under which stream data must be processed have motivated a recent upsurge of on-line classification models [13,7,24,6]. In order to identify changes in time series data generator models, Cavalcante et al [1] have recently proposed a concept drift detector method coined as FEDD. Based on the feature vector similarity given by Pearson correlation distance (or cosine distance), this method monitors the evolution of sequence features in order to test whether a concept change has occurred.…”