東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back to Search results for [ null ]

Switch To: Labeled | MARC Mode | ISBD

Loss functions for binary classifica...

Shen, Yi.

Linked to FindBook

Google Book

Amazon

博客來

Loss functions for binary classification and class probability estimation.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Loss functions for binary classification and class probability estimation./
Author:	Shen, Yi.
Description:	115 p.
Notes:	Source: Dissertation Abstracts International, Volume: 66-06, Section: B, page: 3215.
Contained By:	Dissertation Abstracts International66-06B.
Subject:	Statistics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3179814
ISBN:	0542200554

Loss functions for binary classification and class probability estimation.
Shen, Yi.

Loss functions for binary classification and class probability estimation. - 115 p.

Source: Dissertation Abstracts International, Volume: 66-06, Section: B, page: 3215.

Thesis (Ph.D.)--University of Pennsylvania, 2005.

What are the natural loss functions for binary class probability estimation? This question has a simple answer: so-called "proper scoring rules". These loss functions, known from subjective probability, measure the discrepancy between true probabilities and estimates thereof. They comprise all commonly used loss functions: lob loss, squared error loss, boosting loss (which we derive from boosting's exponential loss), and cost-weighted misclassification losses. We also introduce a larger class of possibly uncalibrated loss functions that can be calibrated with a link function. An example is exponential loss, which is related to boosting.

ISBN: 0542200554Subjects--Topical Terms:

517247
Statistics.

Loss functions for binary classification and class probability estimation.
LDR:02697nmm 2200301 4500 001 1819550
005 20061005085933.5
008 130610s2005 eng d
020 $a 0542200554
035 $a (UnM)AAI3179814
035 $a AAI3179814
040 $a UnM $c UnM
100 1 $a Shen, Yi. $3 1908827
245 1 0 $a Loss functions for binary classification and class probability estimation.
300 $a 115 p.
500 $a Source: Dissertation Abstracts International, Volume: 66-06, Section: B, page: 3215.
500 $a Supervisor: Andreas Buja.
502 $a Thesis (Ph.D.)--University of Pennsylvania, 2005.
520 $a What are the natural loss functions for binary class probability estimation? This question has a simple answer: so-called "proper scoring rules". These loss functions, known from subjective probability, measure the discrepancy between true probabilities and estimates thereof. They comprise all commonly used loss functions: lob loss, squared error loss, boosting loss (which we derive from boosting's exponential loss), and cost-weighted misclassification losses. We also introduce a larger class of possibly uncalibrated loss functions that can be calibrated with a link function. An example is exponential loss, which is related to boosting.
520 $a Proper scoring rules are fully characterized by weight functions o(eta) on class probabilities eta = P[Y = 1]. These weight functions give immediate practical insight into loss functions: high mass of o(eta) points to the class probabilities eta where the proper scoring rule strives for greatest accuracy. For example, both log-loss and boosting loss have poles near zero and one, hence rely on extreme probabilities.
520 $a We show that the freedom of choice among proper scoring rules can be exploited ploited when the two types of misclassification have different costs: one can choose proper scoring rules that focus on the cost c of class 0 misclassification by concentrating o(eta) near c. We also show that cost-weighting uncalibrated loss functions can achieve tailoring. "Tailoring" is often beneficial for classical linear models, whereas non-parametric boosting models show fewer benefits.
520 $a We illustrate "tailoring" with artificial and real datasets both for linear models and for non-parametric models based on trees, and compare it with traditional linear logistic regression and one recent version of boosting, called "LogitBoost".
590 $a School code: 0175.
650 4 $a Statistics. $3 517247
690 $a 0463
710 2 0 $a University of Pennsylvania. $3 1017401
773 0 $t Dissertation Abstracts International $g 66-06B.
790 1 0 $a Buja, Andreas, $e advisor
790 $a 0175
791 $a Ph.D.
792 $a 2005
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3179814