基于证据推理规则的数据分类方法研究

95 浏览量更新于2024-08-27 收藏 489KB PDF 举报

数据分类使用证据推理规则本文总结了数据分类使用证据推理规则的一种新方法。该方法基于 Dempster-Shafer 证据理论（DST），利用 Evidential Reasoning（ER）规则来组合来自不同属性的证据。与传统的 Dempster's combination（DC）规则相比，该方法可以更好地处理不确定性和不确定性。数据分类是数据挖掘和机器学习领域中的一个重要问题。传统的分类方法通常基于概率论和统计学，但是这些方法存在一些缺陷，例如无法处理不确定性和不确定性。在最近的研究中，基于证据理论的分类方法日益受到关注。证据理论可以提供一种更为灵活的方法来处理不确定性和不确定性。 Dempster-Shafer 证据理论（DST）是一种常用的证据理论方法，它可以用于处理不确定性和不确定性。DST 中的证据可以来自不同的属性，每个属性都可以提供一些证据来支持或反对某个分类结果。然而，如何组合来自不同属性的证据是一个挑战性的问题。传统的 Dempster's combination（DC）规则是一种常用的组合规则，但是它存在一些缺陷，例如无法处理不确定性和不确定性。在本文中，我们提出了一个新的组合规则，即 Evidential Reasoning（ER）规则。ER 规则可以更好地处理不确定性和不确定性，并且可以提供一种更为灵活的方法来组合来自不同属性的证据。 ER 规则的基本思想是：首先，计算每个属性的证据权重，然后使用这些权重来组合来自不同属性的证据。与 DC 规则相比，ER 规则可以更好地处理不确定性和不确定性，并且可以提供一种更为灵活的方法来组合来自不同属性的证据。在本文中，我们还提出了一个基于 Sequential Linear Programming（SLP）的优化方法来解决 ER 规则中的优化问题。SLP 方法可以快速地解决优化问题，并且可以提供一种更为精准的解决方案。本文提出了一个基于证据理论的数据分类方法，该方法可以更好地处理不确定性和不确定性，并且可以提供一种更为灵活的方法来组合来自不同属性的证据。该方法可以广泛应用于数据挖掘和机器学习领域。知识点： * 数据分类 * 证据理论 * Dempster-Shafer 证据理论（DST） * Evidential Reasoning（ER）规则 * Dempster's combination（DC）规则 * Sequential Linear Programming（SLP） * 不确定性和不确定性处理 * 证据权重计算 * 优化方法

Knowledge-Based Systems 116 (2017) 144–151

Contents lists available at ScienceDirect

Knowle dge-Base d Systems

journal homepage: www.elsevier.com/locate/knosys

Data classiﬁcation using evidence reasoning rule

Xiaobin Xu

∗

, Jin Zheng

, Jian-bo Yang

, Dong-ling Xu

, Yu-wang Chen

Institute of System Science and Control Engineering, Hangzhou Dianzi University, Hangzhou 310018, Zhejiang, China

Decision and Cognitive Sciences Research Centre, The University of Manchester, Manchester M15 6PB, UK

a r t i c l e i n f o

Article history:

Received 16 June 2016

Revised 21 August 2016

Accepted 1 November 2016

Available online 2 November 2016

Keyword:

Date classiﬁcation

Dempster–Shafer evidence theory (DST)

Evidential reasoning (ER) rule

Reliability and weight of evidence

Sequential linear programming (SLP)

a b s t r a c t

In Dempster–Shafer evidence theory (DST) based classiﬁer design, Dempster’s combination (DC) rule is

commonly used as a multi-attribute classiﬁer to combine evidence collected from different attributes.

The main aim of this paper is to present a classiﬁcation method using a novel combination rule i.e., the

evidence reasoning (ER) rule. As an improvement of the DC rule, the newly proposed ER rule deﬁnes

the reliability and weight of evidence. The former indicates the ability of attribute or its evidence to

provide correct assessment for classiﬁcation problem, and the latter reﬂects the relative importance of

evidence in comparison with other evidence when they need to be combined. The ER rule-based clas-

siﬁcation procedure is expatiated from evidence acquisition and estimation of evidence reliability and

weight to combination of evidence. It is a purely data-driven approach without making any assumptions

about the relationships between attributes and class memberships, and the speciﬁc statistic distributions

of attribute data. Experiential results on ﬁve popular benchmark databases taken from University of Cal-

ifornia Irvine (UCI) machine learning database show high classiﬁcation accuracy that is competitive with

other classical and mainstream classiﬁers.

Introduction

Classiﬁcation problem is one of the most important issues in

data mining and knowledge discovery [1] . Its purpose is to fall a

sample with unknown class into a basket with a label of speciﬁc

class where an appropriate classiﬁer should be used to analyze the

attributes of this sample. Classiﬁcation are fundamental to many

theoretical and practical applications, including pattern recogni-

tion [2–4] , fault diagnosis [5–7] , and image processing [8–10] , etc.

Many well-known methods have been proposed to solve classiﬁca-

tion problems, including k nearest neighbors (k-NN) [11] , support

vector machine [12] , naive Bayes [13] , Bayes net [14] , decision tree

learner [15] , random forest [16] , and other latest techniques, such

as gravitational inspired classiﬁer [17] , feature vector graph-based

classiﬁer [18] , and Learning automata(LA)-based classiﬁer [19] , and

so on.

From the perspective of uncertain information processing, the

imprecision or even incorrectness of classiﬁcation results is likely

to be caused by the fact that the values of attributes of a sam-

ple cannot categorically point to a certain class, that is to say, the

boundaries of attributes among different classes are commonly im-

precise, or even overlapping [20–22] . As a result, Dempster–Shafer

evidence theory (DST) can provide an available mechanism to deal

∗

Corresponding author.

E-mail address: xuxiaobin1980@163.com (X. Xu).

with the classiﬁcation imprecision. In detail, a frame of discern-

ment (FoD) needs to be ﬁrstly determined which includes all pre-

assigned class memberships. The next step is to obtain a basic be-

lief assignment (BBA), i.e., a belief distribution (BD) function, in

which the belief degrees are used to measure the extents to which

the attributes of a sample supports each class and the subsets of

the classes. Such a BBA or a BD can be also named as a piece of ev-

idence. There are different ways for generating BBAs from different

types of attribute information. The typical ways include core sam-

ple [20] , neural network [21] , k-NN [22] , expert system [23] and

so on. The ﬁnal step is to use Dempster’s combination (DC) rule to

fuse these BBAs and then make a classiﬁcation decision according

to the fused results. The aim of combination is to reduce the clas-

siﬁcation imprecision by fusing multi-source attribute information.

Recently, the evidential reasoning (ER) rule has been estab-

lished to advance the seminal Dempster–Shafer evidence theory

[24–28] and the original ER algorithm [29–32] . Compared with the

DC rule, the main advance of the ER rule is to propose a novel con-

cept of weighted evidence (WE) and extend to WE with Reliability

(WER) in order to characterize evidence in complement of BBA or

BD introduced in the DST. As a result, the implementation of the

orthogonal sum operation on WEs and WERs leads to the estab-

lishment of the new ER rule [33] . The most important property of

the ER rule is that it constitutes a generic conjunctive probabilis-

tic reasoning process, or a generalized Bayesian inference process

which can be implemented on the power set of FoD. It has been

http://dx.doi.org/10.1016/j.knosys.2016.11.001

下载后可阅读完整内容，剩余7页未读，立即下载

weixin_38618315

粉丝: 1
资源: 921

基于证据推理规则的数据分类方法研究

vechicle classification using convolutional neural networks

Data ClassifiCation Algorithms andApplications 2014.英文版

Belief rule based expert system for classification problems with new rule activation and weight calculation procedures

Interval-valued analysis for discriminative gene selection and tissue sample classification using microarray data

Cancer Classification Using Collaborative Representation Classifier Based on Non-convex Lp-norm and Novel Decision Rule

Classification Using PNARM-开源

Pattern Classification Using Ensemble Methods

ClassificationDataset.rar_SVM data_data classification_data set_

Color Texture Classification Using Wavelet Transform

Automatic Modulation Classification Using Deep Learning

最新资源