Limits on the majority vote accuracy in classifier fusion

By Ludmila I. Kuncheva, Christopher J. Whitaker, Catherine A. Shipp, Robert P. W. Duin. Pattern Analysis and Applications, 2003

408 citations recorded by the source corpus.

Subfields: Economics and Econometrics, Political Science and International Relations

View Paper: https://www.semanticscholar.org/paper/42f7cfb58bf26ee7c7d414bb848e6b27248868db

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Economics and Econometrics

In Political Science and International Relations

Citations in the corpus

Ensemble Learning (2019, neutral): “Leistner et al. (2009) proposed a novel boosting algorithm, On-line GradientBoost, which outperformed On-line AdaBoost on standard machine learning problems and common computer vision applications.”
AggNet: Deep Learning From Crowds for Mitosis Detection in Breast Cancer Histology Images (2016, methodology): “5) Aggregated Labels (AL): To investigate the aggregated labels of the crowdvotes matrix , we first run majority voting (MV) [29] and GLAD [35] methods on without any quality control, referred to as MV-NoQ and GLAD-NoQ respectively.”
EUSBoost: Enhancing ensembles for highly imbalanced data-sets by evolutionary undersampling (2013, methodology): “One of the most common is the Q -statistic [58], which has been widely applied in classi ﬁ er ensembles [32,52].”
Crowdsourced Data Management: A Survey (2016, background): “As a single worker may be biased for some tasks, most existing works [15], [22], [58], [69], [74], [138], [139] often employ a quality-control strategy, called answer aggregation (or voting strategy).”
A Survey on Ensemble Learning for Data Stream Classification (2017, methodology): “Finally, some authors focus on determining the limits of majority voting [Kuncheva et al. 2003] or on comparing multiple voting methods using a probabilistic framework [Kuncheva and Rodrı́guez 2014], yet conclusions obtained are often limited to specific cases.”
Diversity creation methods: a survey and categorisation (2004, background): “The vast majority of empirical evidence examining classier diversity is due to Kuncheva [ 24 , 25, 20, 1, 26, 27, 28, 29].”
Ensemble deep learning: A review (2021, background): “In majority voting, the analysis in [125] showed that the pairwise dependence among the base learners plays an important role and for the classification of images, the prediction of shallow networks is more diverse as compared to the deeper networks [126].”
Relationships between combination methods and measures of diversity in combining classifiers (2002, background): “In a previous study, we proved that there is a functional relationship between the Q statistic and the upper and the lower limits of the majority vote accuracy [ 14 ].”
Measures of Diversity in Classifier Ensembles and Their Relationship with the Ensemble Accuracy (2003, methodology): “The relationship between Q av and the limits of P maj has been formally derived (Kuncheva et al., 2000), and for these limit cases, Q av can take values ?”
The Relative Performance of Ensemble Methods with Deep Convolutional Neural Networks for Image Classification (2017, background): “[Kuncheva et al., 2003] showed pairwise dependence plays an an important role in majority voting.”
An experimental comparison of performance measures for classification (2009, background): “…many works have been devoted to study the resulting accuracy of an ensemble of combined classifiers given the original accuracies and some other conditions (Melnik et al., 2004; Narasimhamurthy, 2005; Kuncheva et al., 2003) or how to combine them in order to increase accuracy (Young Sohn, 1999).”
Classifier selection for majority voting (2005, background): “Then the best classifier for each cluster is identified and selected to classify the new pattern if it falls into its region [18,24].”
A survey of multiple classifier systems as hybrid systems (2014, background): “Pairwise measures averaging a measure between each classifier pair in an ensemble, such as Q-statistic [58], kappa-statistics [63], disagreement [64] and double-fault measure [61,65].”