Calculating the relative importance of condition attributes based on the characteristics of decision rules and attribute reducts: Application to crowdfunding
Calculating the relative importance of condition attributes based on the characteristics of decision rules and attribute reducts: Application to crowdfunding
复制标题
DOI：
10.1016/j.ejor.2020.03.039
复制
发表时间：
2020-10-16
影响因子：
6.4
通讯作者：
Ford, Liz
中科院分区：
文献类型：
--
作者：
Chakhar, Salem;Ishizaka, Alessio;Ford, Liz
众筹是一种通过筹集大量资金来资助项目或企业的做法，通常是通过互联网。Lendwith CARE是首批众筹平台之一，专门致力于通过伙伴小额金融机构支持发展中国家的个人和团体企业家。Lendwith CARE的一个关键目标是确定在做出投资决策时影响投资者/潜在投资者的属性(即众筹项目在其在线描述中的特征)。本文提出了一种基于决策规则的方法来解决这一问题。这种方法依赖于基于优势的粗糙方法(DRSA)，这是一种著名的多准则排序方法。DRSA的输出是IF-THEN决策规则的集合和属性约简的集合。根据决策规则和属性约简的特点，提出了计算条件属性相对重要性的新方法。基于决策规则的度量被参数化，以便使用学习和测试数据集来考虑决策规则的特征。建议的度量可以聚合为一个综合度量，表示每个条件属性的总体重要性。此外，为了计算条件属性集合在一起的相对重要性，对所提出的度量进行了扩展。此外，扩展了基于决策规则的度量来评估条件属性特定值的相对重要性。所提出的方法已经使用来自Lendthcare的真实世界数据进行了应用和验证。(C)2020爱思唯尔B.V.保留所有权利。
Crowdfunding is the practice of funding a project or venture by raising monetary contributions from a large number of people, typically via the Internet. Lendwithcare is amongst the first crowdfunding platforms specifically dedicated to support individual and group entrepreneurs in developing countries through partner microfinance institutions. A key objective of Lendwithcare is to identify the attributes (i.e., the characteristics of crowdfunding projects in their online descriptions) that affect investors/potential investors when taking their investment decision. This paper proposes a decision rule-based approach to address this issue. This approach relies on the Dominance-Based Rough Approach (DRSA), which is a well-known multicriteria sorting method. The outputs of DRSA are a collection of if-then decision rules and a collection of attribute reducts. In this paper, new measures are proposed for calculating the relative importance of condition attributes based on the characteristics of decision rules and of attribute reducts. Decision rule-based measures are parameterised in order to consider the characteristics of decision rules using both learning and testing datasets. The proposed measures can be aggregated into a comprehensive measure indicating the overall importance of each condition attribute. Furthermore, the proposed measures are extended in order to compute the relative importance of a collection of condition attributes taken together. In addition, decision rule-based measures are extended to evaluate the relative importance of specific values of condition attributes. The proposed approach has been applied and validated using real-world data from Lendwithcare. (C) 2020 Elsevier B.V. All rights reserved.