Learning classifiers from only positive and unlabeled data
Published 24 August 2008
Charles Elkan, Keith Noto
Citations1,034
Generate an AI Snapshot to get a quick, structured summary of this paper.
Study Snapshot
ObjectiveStudy objective
MethodsResearch methodology
PopulationPopulation studied
Sample sizeSample sizes
OutcomesStudy outcomes here
ResultsStudy results comes here
LimitationsResearch study limitations comes here
A concise AI-generated summary of the paper will appear here once you click Generate AI Snapshot.
TL;DR
This paper shows that models trained using the new methods perform better than the current state-of-the-art biased SVM method for learning from positive and unlabeled examples, and applies them to solve a real-world problem: identifying protein records that should be included in an incomplete specialized molecular biology database.
Abstract
The input to an algorithm that learns a binary classifier normally consists of two sets of examples, where one set consists of positive examples of the concept to be learned, and the other set consists of negative examples. However, it is often the case that the available training data are an incomplete set of positive examples, and a set of unlabeled examples, some of which are positive and some of which are negative. The problem solved in this paper is how to learn a standard binary classifier given a nontraditional training set of this nature.
Keywords
Computer Science
Journal of the American Statistical AssociationStatistical Analysis With Missing Data
17,489 Citations1989Maureen Lahiff, Roderick J. A. Little +1 more
Wiley series in probability and statisticsStatistical Analysis with Missing Data
8,363 Citations2002Roderick J. A. Little, Donald B. Rubin
Neural ComputationEstimating the Support of a High-Dimensional Distribution
5,946 Citations2001Bernhard Schölkopf, John Platt +3 more
The algorithm is a natural extension of the support vector algorithm to the case of unlabeled data by carrying out sequential optimization over pairs of input patterns and providing a theoretical analysis of the statistical performance of the algorithm.
Nucleic Acids ResearchThe SWISS-PROT protein knowledgebase and its supplement TrEMBL in 2003
3,996 Citations2003Brigitte Boeckmann
The SWISS-PROT protein knowledgebase connects amino acid sequences with the current knowledge in the Life Sciences by providing an interdisciplinary overview of relevant information by bringing together experimental results, computed features and sometimes even contradictory conclusions.
Machine LearningSupport Vector Data Description
3,484 Citations2003David M. J. Tax, Robert P. W. Duin
The Support Vector Data Description (SVDD) is presented which obtains a spherically shaped boundary around a dataset and analogous to the Support Vector Classifier it can be made flexible by using other kernel functions.
One-class svms for document classification
1,225 Citations2002Larry M. Manevitz, Malik Yousef
The SVM approach as represented by Schoelkopf was superior to all the methods except the neural network one, where it was, although occasionally worse, essentially comparable.
Transforming classifier scores into accurate multiclass probability estimates
1,099 Citations2002Bianca Zadrozny, Charles Elkan
This work shows how to obtain accurate probability estimates for multiclass problems by combining calibrated binary probability estimates, and proposes a new method for obtaining calibrated two-class probability estimates that can be applied to any classifier that produces a ranking of examples.
The MIT Press eBooksProbabilities for SV Machines
1,015 Citations2000John C. Platt
This chapter contains sections titled: Introduction, Fitting a Sigmoid After the SVM, Empirical Tests, Conclusions, Appendix: Pseudo-code for the Sigmoids Training.
Machine LearningA note on Platt’s probabilistic outputs for support vector machines
868 Citations2007Hsuan-Tien Lin, Chih‐Jen Lin +1 more
An improved algorithm that theoretically converges and avoids numerical difficulties is proposed for Platt’s probabilistic outputs for Support Vector Machines.
Nucleic Acids ResearchTCDB: the Transporter Classification Database for membrane transport protein analyses and information
863 Citations2005Milton H. Saier
The Transporter Classification Database (TCDB) is a web accessible, curated, relational database containing sequence, classification, structural, functional and evolutionary information about transport systems from a variety of living organisms.
Building text classifiers using positive and unlabeled examples
672 Citations2004B. Liu, Y. Dai +3 more
A more principled approach to solving the problem of building text classifiers using positive and unlabeled examples based on a biased formulation of SVM is proposed, and it is shown experimentally that it is more accurate than the existing techniques.
Autonomously semantifying wikipedia
381 Citations2007Fei Wu, Daniel S. Weld
This paper argues that autonomously "Semantifying Wikipedia" is the best way to solve the chicken-and-egg problem, and describes a prototype implementation of a self-supervised, machine learning system which realizes the vision.
Learning with positive and unlabeled examples using weighted logistic regression
334 Citations2003Wee Sun Lee, Bing Liu
A performance measure that can be estimated from positive and unlabeled examples for evaluating retrieval performance, which is proportional to the product of precision and recall, can be used with a validation set to select regularization parameters for logistic regression.
BiometricsPresence‐Only Data and the EM Algorithm
305 Citations2008Gillian Ward, Trevor Hastie +3 more
An expectation-maximization algorithm is proposed to estimate the underlying presence-absence logistic model for presence-only data and it is shown that the population prevalence of a species is only identifiable when there is some unrealistic constraint on the structure of theLogistic model.
IEEE Transactions on Knowledge and Data EngineeringPebl:web page classification without negative examples
255 Citations2004Hwanjo Yu, Jiawei Han +1 more
The paper presents a framework, called positive example based learning (PEBL), for Web page classification which eliminates the need for manually collecting negative training examples in preprocessing and applies an algorithm, called mapping-convergence (M-C), to achieve high classification accuracy as high as that of a traditional SVM.
Nucleic Acids ResearchThe Molecular Biology Database Collection: 2005 update
231 Citations2004Michael Y. Galperin
The 2005 update includes 719 databases, 171 more than the 2004 one, and is organized in a hierarchical classification that simplifies the process of finding the right database for any given task.
Theoretical Computer ScienceLearning from positive and unlabeled examples
225 Citations2005François Denis, Rémi Gilleron +1 more
It is proved that any class learnable in the statistical query learning model is learnable from positive statistical queries and instance statistical queries only if a lower bound on the weight of any, target concept f can be estimated in polynomial time.
IEEE Transactions on Knowledge and Data EngineeringText classification without negative examples revisit
220 Citations2006Gabriel Pui Cheong Fung, Jeffrey Xu Yu +2 more
A labeling heuristic called PNLH is proposed, which aims at extracting high quality positive examples and negative examples from U and can be used on top of any existing classifiers and is highly feasible, especially in the situation where |P| is extremely small.
Lecture notes in computer sciencePAC Learning from Positive Statistical Queries
177 Citations1998François Denis
This work defines a PAC learning model from positive and unlabeled examples, and defines a PAC learning model from positive and unlabeled statistical queries and shows that k-DNF and k-decision lists are learnable in both models.
Nucleic Acids ResearchThe Molecular Biology Database Collection: 2008 update
130 Citations2007Michael Y. Galperin
The 2008 update includes 1078 databases, 110 more than the previous one, and the links to more than 80 databases have been updated and 25 obsolete databases have be removed from the list.
BioinformaticsPSoL: a positive sample only learning algorithm for finding non-coding RNA genes
99 Citations2006Chunlin Wang, Chris Ding +2 more
A new approach called positive sample only learning (PSoL) to predict ncRNA genes in the Escherichia coli genome is described, which shows that PSoL can achieve about 80% accuracy in recovery of known ncRNAs.
International Conference Information ProcessingText Classification from Positive and Unlabeled Examples
82 Citations2002François Denis, Rémi Gilleron +1 more
Experimental results show that performan e of the naive Bayes algorithm for learning from positive and unlabeled examples is omparable with naive Bayesian algorithm forLearning from labeled data.
Machine LearningSingle-Class Classification with Mapping Convergence
66 Citations2005Hwanjo Yu
This work presents an SCC algorithm called Mapping Convergence (MC) that computes an accurate boundary of the target class from positive and unlabeled data (without labeled negative data) and presents Support Vector Mapping convergence (SVMC) which optimizes the MC algorithm for fast training.
Making generative classifiers robust to selection bias
29 Citations2007Andrew T. Smith, Charles Elkan
A learning method based on the expectation maximization (EM) algorithm that, while not always able to overcome arbitrary labeling bias, learns SMMs with higher test-set accuracy in real-world data sets (with MNAR bias) than existing learning methods that are proven to overcome MAR bias.
A Bayesian network framework for reject inference
28 Citations2004Andrew T. Smith, Charles Elkan
A systematic analysis of different cases that arise in reject inference, using Bayesian networks to formalize each case as a set of conditional independence relationships and identify eight cases, including the familiar missing completely at random, missing at random (MAR), and missing not at Random (MNAR) cases.
Lecture notes in computer sciencePartially Supervised Classification – Based on Weighted Unlabeled Samples Support Vector Machine
18 Citations2005Zhigang Liu, Wenzhong Shi +2 more
Experimental results indicate that the proposed PSC method is more robust than 1-SVM and has comparable accuracy to a standard SVM.
Lecture notes in computer scienceFinding Transport Proteins in a General Protein Database
13 Citations2007Sanmay Das, Milton H. Saier +1 more
A maximum-entropy classifier is described, trained on SwissProt records, that achieves high precision and recall in cross-validation experiments and is deployed as part of a pipeline for updating TCDB, the Transport Classification Database.