On equivalence relationships between classification and ranking algorithms
We demonstrate that there are machine learning algorithms that can achieve success for two separate tasks simultaneously, namely the tasks of classification and bipartite ranking. This means that advantages gained from solving one task can be carried over to the other task, such as the ability to...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | en_US |
| Published: |
Association for Computing Machinery
2012
|
| Online Access: | http://hdl.handle.net/1721.1/75726 https://orcid.org/0000-0001-6541-1650 |
| _version_ | 1826193368917475328 |
|---|---|
| author | Ertekin, Seyda Rudin, Cynthia |
| author2 | Sloan School of Management |
| author_facet | Sloan School of Management Ertekin, Seyda Rudin, Cynthia |
| author_sort | Ertekin, Seyda |
| collection | MIT |
| description | We demonstrate that there are machine learning algorithms that can achieve success for two separate
tasks simultaneously, namely the tasks of classification and bipartite ranking. This means that
advantages gained from solving one task can be carried over to the other task, such as the ability
to obtain conditional density estimates, and an order-of-magnitude reduction in computational
time for training the algorithm. It also means that some algorithms are robust to the choice of
evaluation metric used; they can theoretically perform well when performance is measured either
by a misclassification error or by a statistic of the ROC curve (such as the area under the curve).
Specifically, we provide such an equivalence relationship between a generalization of Freund et
al.’s RankBoost algorithm, called the “P-Norm Push,” and a particular cost-sensitive classification
algorithm that generalizes AdaBoost, which we call “P-Classification.”We discuss and validate the
potential benefits of this equivalence relationship, and perform controlled experiments to understand P-Classification’s empirical performance. There is no established equivalence relationship for logistic regression and its ranking counterpart, so we introduce a logistic-regression-style algorithm that aims in between classification and ranking, and has promising experimental performance with respect to both tasks. |
| first_indexed | 2024-09-23T09:37:58Z |
| format | Article |
| id | mit-1721.1/75726 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T09:37:58Z |
| publishDate | 2012 |
| publisher | Association for Computing Machinery |
| record_format | dspace |
| spelling | mit-1721.1/757262022-09-26T12:45:18Z On equivalence relationships between classification and ranking algorithms Ertekin, Seyda Rudin, Cynthia Sloan School of Management Ertekin, Seyda Rudin, Cynthia We demonstrate that there are machine learning algorithms that can achieve success for two separate tasks simultaneously, namely the tasks of classification and bipartite ranking. This means that advantages gained from solving one task can be carried over to the other task, such as the ability to obtain conditional density estimates, and an order-of-magnitude reduction in computational time for training the algorithm. It also means that some algorithms are robust to the choice of evaluation metric used; they can theoretically perform well when performance is measured either by a misclassification error or by a statistic of the ROC curve (such as the area under the curve). Specifically, we provide such an equivalence relationship between a generalization of Freund et al.’s RankBoost algorithm, called the “P-Norm Push,” and a particular cost-sensitive classification algorithm that generalizes AdaBoost, which we call “P-Classification.”We discuss and validate the potential benefits of this equivalence relationship, and perform controlled experiments to understand P-Classification’s empirical performance. There is no established equivalence relationship for logistic regression and its ranking counterpart, so we introduce a logistic-regression-style algorithm that aims in between classification and ranking, and has promising experimental performance with respect to both tasks. National Science Foundation (U.S.) (Grant no. IIS-1053407) Massachusetts Institute of Technology. Energy Initiative 2012-12-13T20:37:17Z 2012-12-13T20:37:17Z 2011-10 2011-08 Article http://purl.org/eprint/type/JournalArticle 1532-4435 1533-7928 http://hdl.handle.net/1721.1/75726 Ertekin, Seyda and Cynthia Rudin. "On Equivalence Relationships Between Classification and Ranking Algorithms." Journal of Machine Learning Research 12 (2011) 2905-2929. https://orcid.org/0000-0001-6541-1650 en_US http://jmlr.csail.mit.edu/papers/volume12/ertekin11a/ertekin11a.pdf Journal of Machine Learning Research Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Association for Computing Machinery MIT Press |
| spellingShingle | Ertekin, Seyda Rudin, Cynthia On equivalence relationships between classification and ranking algorithms |
| title | On equivalence relationships between classification and ranking algorithms |
| title_full | On equivalence relationships between classification and ranking algorithms |
| title_fullStr | On equivalence relationships between classification and ranking algorithms |
| title_full_unstemmed | On equivalence relationships between classification and ranking algorithms |
| title_short | On equivalence relationships between classification and ranking algorithms |
| title_sort | on equivalence relationships between classification and ranking algorithms |
| url | http://hdl.handle.net/1721.1/75726 https://orcid.org/0000-0001-6541-1650 |
| work_keys_str_mv | AT ertekinseyda onequivalencerelationshipsbetweenclassificationandrankingalgorithms AT rudincynthia onequivalencerelationshipsbetweenclassificationandrankingalgorithms |