A unifying view on dataset shift in classification

Moreno-Torres, Jose G.; Raeder, Troy; Aláiz-Rodríguez, Rocío; Chawla, Nitesh V.; Herrera, F.

doi:10.1016/j.patcog.2011.06.019

Cited by 697 publications

(432 citation statements)

References 46 publications

Supporting

Mentioning

392

Contrasting

Unclassified

Order By: Relevance

“…2b, we see the MAR-C assumption, where S is a function of the true (unobservable) class label. In the missing data literature (Little and Rubin 2002), this scenario is classified as missing-not-at-random, and, since the missingness depends only in the class, Moreno-Torres et al (2012) name it as missing-completely-at-random class dependent (MAR-C). We can connect the two semi-supervised scenarios further, with the following observation: MCAR can be seen as a special case of MAR-C.…”

Section: Class-prior-change Semi-supervised Scenario (Or Mar-c)mentioning

confidence: 99%

“…For example another assumption used in the semi-supervised learning is when the missingness mechanism depends directly only on the features or in other words the labelling of an example is conditionally independent of the class, given the feature values. This scenario is known in missing data literature as missing at random (MAR) (Moreno-Torres et al 2012). The importance of this assumption is also presented in the framework of semi-supervised regression by Lafferty and Wasserman (2007).…”

Section: Future Workmentioning

confidence: 99%

“…In terms of data, we tackle two semi-supervised scenarios-when the labels are missing completely at random (MCAR), and a missing-not-at-random scenario (MAR-C) where the class labels are missing according to a mechanism, dependent on the class label itself (Moreno-Torres et al 2012). The latter might occur for example when there is a social stigma associated with reporting of a label, such as income levels or HIV incidence.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Simple strategies for semi-supervised feature selection

Sechidis

Brown

2017

Mach Learn

View full text Add to dashboard Cite

What is the simplest thing you can do to solve a problem? In the context of semisupervised feature selection, we tackle exactly this-how much we can gain from two simple classifier-independent strategies. If we have some binary labelled data and some unlabelled, we could assume the unlabelled data are all positives, or assume them all negatives. These minimalist, seemingly naive, approaches have not previously been studied in depth. However, with theoretical and empirical studies, we show they provide powerful results for feature selection, via hypothesis testing and feature ranking. Combining them with some "soft" prior knowledge of the domain, we derive two novel algorithms (Semi-JMI, Semi-IAMB) that outperform significantly more complex competing methods, showing particularly good performance when the labels are missing-not-at-random. We conclude that simple approaches to this problem can work surprisingly well, and in many situations we can provably recover the exact feature selection dynamics, as if we had labelled the entire dataset.

show abstract

Section: Class-prior-change Semi-supervised Scenario (Or Mar-c)mentioning

confidence: 99%

Section: Future Workmentioning

confidence: 99%

mentioning

confidence: 99%

See 1 more Smart Citation

Simple strategies for semi-supervised feature selection

Sechidis

Brown

2017

Mach Learn

View full text Add to dashboard Cite

show abstract

“…Distribution shift: The most obvious type of context change is given by a change in the data distribution. One common way of looking at a change in the data is known as data shift [53,49]. Data shift is usually classified into covariate shift, prior probability shift and concept drift, but more thorough classifications have been developed (see, e.g., [49]).…”

Section: Taxonomy Of Context Changesmentioning

confidence: 99%

Reframing in context: A systematic approach for model reuse in machine learning

Hernández-Orallo

Martínez-Usó

Prudêncio

et al. 2016

AIC

View full text Add to dashboard Cite

“…Note that while we have given a brief overview of concept drift in particular, Moreno-Torres et al [64] provide a more focused overview of different forms of "dataset drift", as well as their causes.…”

Section: Concept Driftmentioning

confidence: 99%

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

2012

Self Cite

View full text Add to dashboard Cite

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.

show abstract

A unifying view on dataset shift in classification

Cited by 697 publications

References 46 publications

Simple strategies for semi-supervised feature selection

Simple strategies for semi-supervised feature selection

Reframing in context: A systematic approach for model reuse in machine learning

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

Contact Info

Product

Resources

About