Learning from streaming data with concept drift and imbalance: an overview

Abstract

The primary focus of machine learning has traditionally been on learning from data assumed to be sufficient and representative of the underlying fixed, yet unknown, distribution. Such restrictions on the problem domain paved the way for development of elegant algorithms with theoretically provable performance guarantees. As is often the case, however, real-world problems rarely fit neatly into such restricted models. For instance class distributions are often skewed, resulting in the "class imbalance" problem. Data drawn from non-stationary distributions is also common in real-world applications, resulting in the "concept drift" or "non-stationary learning" problem which is often associated with streaming data scenarios. Recently, these problems have independently experienced increased research attention, however, the combined problem of addressing all of the above mentioned issues has enjoyed relatively little research. If the ultimate goal of intelligent machine learning algorithms is to be able to address a wide spectrum of real-world scenarios, then the need for a general framework for learning from, and adapting to, a non-stationary environment that may introduce imbalanced data can be hardly overstated. In this paper, we first present an overview of each of these challenging areas, followed by a comprehensive review of recent research for developing such a general framework.

Learning from imbalanced data: open challenges and future directions

“…Dynamical nature of data that arrive either in batches or online poses new challenges when imbalanced distributions are to be expected [26]. Whether dealing with stationary or evolving streams we require adaptive methods that are able to deal with skewed objects coming in real time.…”

Integrative methods for analyzing big data in precision medicine

“…One of the possible future directions in addressing this challenge is the utilisation of so-called ``anytime algorithms'' [170] that can learn from streaming data (e.g., time-dependent Bayesian classifiers) [171] and that still return a valuable result if their execution is interrupted at any time. Moreover, in the future, we will have access to more and more time series data.…”

Sliding Reservoir Approach for Delayed Labeling in Streaming Data Classification

“…When concept drifts occur in a data stream, certain amount of labeled data samples are needed for training new supervised or semi-supervised learning models [3]. A request for labels on selected data samples will be made prior to the training.…”

“…Examples of concept drifting data streams are weather data stream, financial data stream, and online-opinion data stream. Concept drifting data streams require the data mining framework to be able to detect changes in the stream, and adapt to them so that the learning model is kept upto-date [3]. Numerous studies have been done on designing such adapting data mining frameworks [4][5][6][7][8][9][10][11].…”