Classification in Presence of Drift and Latency

Krempl, Georg; Hofer, Vera

doi:10.1109/icdmw.2011.47

Cited by 26 publications

(30 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In these approaches, each class is represented as a mixture of subpopulations, and distributions of the unlabeled data from subpopulations are tracked and matched to those known subpopulations based on initial labeled data. Special cases where subpopulations are mixtures of fixed number of Gaussians is described in [46], and for cases, when drift is only due to the changes in priors of the subpopulations is addressed in [47] and [48]. Another example of this group of algorithms is Krempl's APT algorithm [49].…”

Section: E Nonstationary Environments With Verification Latencymentioning

confidence: 99%

COMPOSE: A Semisupervised Learning Framework for Initially Labeled Nonstationary Streaming Data

Dyer

Capo

Polikar

2014

IEEE Trans. Neural Netw. Learning Syst.

141

120

View full text Add to dashboard Cite

Abstract-An increasing number of real-world applications are associated with streaming data drawn from drifting and nonstationary distributions that change over time. These applications demand new algorithms that can learn and adapt to such changes, also known as concept drift. Proper characterization of such data with existing approaches typically requires substantial amount of labeled instances, which may be difficult, expensive, or even impractical to obtain. In this paper, we introduce compacted object sample extraction (COMPOSE), a computational geometry-based framework to learn from nonstationary streaming data, where labels are unavailable (or presented very sporadically) after initialization. We introduce the algorithm in detail, and discuss its results and performances on several synthetic and real-world data sets, which demonstrate the ability of the algorithm to learn under several different scenarios of initially labeled streaming environments. On carefully designed synthetic data sets, we compare the performance of COMPOSE against the optimal Bayes classifier, as well as the arbitrary subpopulation tracker algorithm, which addresses a similar environment referred to as extreme verification latency. Furthermore, using the real-world National Oceanic and Atmospheric Administration weather data set, we demonstrate that COMPOSE is competitive even with a well-established and fully supervised nonstationary learning algorithm that receives labeled data in every batch.Index Terms-Alpha shape, concept drift, nonstationary environment, semisupervised learning (SSL), verification latency.

show abstract

Section: E Nonstationary Environments With Verification Latencymentioning

confidence: 99%

COMPOSE: A Semisupervised Learning Framework for Initially Labeled Nonstationary Streaming Data

Dyer

Capo

Polikar

2014

IEEE Trans. Neural Netw. Learning Syst.

141

120

View full text Add to dashboard Cite

show abstract

“…First, boosting could be carried out on the semi-parametric GAM to see if this produces further performance gains (Bühlmann and Hothorn, 2007;Tutz and Binder, 2008). Second, using a different type of GAM may offer alternative A C C E P T E D M A N U S C R I P T ways to handle class imbalance (Calabrese and Osmetti, 2013 Finally, exploring how population drift may affect model performance would also be an interesting area of research (Krempl and Hofer, 2011). More practically, testing over various prediction horizons (18, 24 months) and perhaps fitting models to a longer time span than the one used in this study would be beneficial before deployment either within financial institutions or by regulatory This section contains results of testing for differences in performance across classifiers trained using the AUC and referred to in section 5.1 of the main text.…”

mentioning

confidence: 99%

An empirical comparison of classification algorithms for mortgage default prediction: evidence from a distressed mortgage market

Fitzpatrick

Mues

2016

European Journal of Operational Research

View full text Add to dashboard Cite

Highlights• We evaluate default prediction performance of machine learning/regression models.• Including boosted trees, random forests, penalised linear/semi-parametric logistic regression.• Using data on over 300,000 residential mortgage loans.• The results indicate varying degrees of predictive power.• Statistical tests suggest boosted regression trees outperform penalised logistic regression.

show abstract

“…In contrast to Biernacki et al (2002) and Beninel et al (2012), it is assumed that the population contains several subpopulations, and that these components do not necessarily behave in the same way as the whole class to which they belong. In contrast to the model in Krempl and Hofer (2011), the number of subpopulations can change over the course of time. The method is based on the idea that unlabelled data give information on the new locations of subpopulations.…”

Section: Introductionmentioning

confidence: 81%

“…Other definitions (see for example Ditzler et al, 2012;Krempl & Hofer, 2011), have a strong focus on the time component, i.e. changes are distinguished in terms of whether they are gradual (i.e.…”

Section: Introductionmentioning

confidence: 99%

“…In addition, the method only applies to continuous features. Krempl and Hofer (2011) proposed a method for only local changes where the parameters were estimated by means of an EM algorithm. Some of the disadvantages of this model are that the estimation of particular weights needs labelled data (for a discussion and the use of auxiliary techniques see Krempl & Hofer, 2011), and that the number of components in the mixture distribution is fixed.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Classification in Presence of Drift and Latency

Cited by 26 publications

References 13 publications

COMPOSE: A Semisupervised Learning Framework for Initially Labeled Nonstationary Streaming Data

COMPOSE: A Semisupervised Learning Framework for Initially Labeled Nonstationary Streaming Data

An empirical comparison of classification algorithms for mortgage default prediction: evidence from a distressed mortgage market

Adapting a classification rule to local and global shift when only unlabelled data are available

Contact Info

Product

Resources

About