Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


masters_thesis.pdf (1.79 MB)

ETD Abstract Container

Abstract Header

Why Boosting Works: Analyses for Noisy Classification and Ranking Problems

Liu, Rui, Liu
http://rave.ohiolink.edu/etdc/view?acc_num=case1474223230237483

Abstract Details

2017, Master of Sciences, Case Western Reserve University, EECS - Computer and Information Sciences.
In this thesis, I study Boosting algorithms, which are a family of machine learning algorithms that aggregate base predictors (e.g. classifiers or rankers) into an accurate final predictor. I theoretically analyze the behavior of such algorithms in the context of two problems: ranking and classification with mislabeling noise. In the context of classification with mislabeling noise, I prove that AdaBoost with linear learner as base leaner is able to perfectly recover the zero-error concept with respect to true labels after certain boosting rounds in the presence of one-sided noise, under some ideal assumptions. I empirically verify the theoretical conclusions of my analysis on synthetic datasets. Experiments on real-world datasets with one-sided noise are also performed and their results broadly support my analysis. In the context of ranking, I analyze the behavior of previously proposed RankBoost algorithm. I show that RankBoost suffers from several flaws including the violation of its desired theoret- ical property in certain scenarios. I then propose a modification to RankBoost, that we call CrankBoost (Corrected RankBoost), that does in fact have the de- sired theoretical properties. I empirically validate that CrankBoost outperforms RankBoost on real ranking datasets.
Soumya Ray (Advisor)
Harold Connamacher (Committee Member)
Vincenzo Liberatore (Committee Member)
133 p.
Computer Science
Boosting; Noise; Classification; Ranking; AdaBoost; RankBoost

Recommended Citations

Citations

  • Liu, Liu, R. (2017). Why Boosting Works: Analyses for Noisy Classification and Ranking Problems [Master's thesis, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1474223230237483

    APA Style (7th edition)

  • Liu, Liu, Rui. Why Boosting Works: Analyses for Noisy Classification and Ranking Problems. 2017. Case Western Reserve University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1474223230237483.

    MLA Style (8th edition)

  • Liu, Liu, Rui. "Why Boosting Works: Analyses for Noisy Classification and Ranking Problems." Master's thesis, Case Western Reserve University, 2017. http://rave.ohiolink.edu/etdc/view?acc_num=case1474223230237483

    Chicago Manual of Style (17th edition)

case1474223230237483
503
© 2016, some rights reserved.Creative Commons License Why Boosting Works: Analyses for Noisy Classification and Ranking Problems by Rui Liu Liu is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.