Rule Extraction from Support Vector Machine Using Modified Active Learning Based Approach: An Application to CRM

被引：0

作者：

Farquad, M. A. H. ^{[1
,2
]}

Ravi, V. ^{[1
]}

Raju, S. Bapi ^{[2
]}

机构：

[1] Inst Dev & Res Banking Technol, Castle Hills Rd 1, Hyderabad 500057, Andhra Pradesh, India

[2] Univ Hyderabad, Dept Comput & Informat Sci, Hyderabad 500046, Andhra Pradesh, India

来源：

KNOWLEDGE-BASED AND INTELLIGENT INFORMATION AND ENGINEERING SYSTEMS, PT I | 2010年 / 6276卷

关键词：

Support Vector Machines; Rule Extraction; modified active learning based approach; Customer Churn; CUSTOMER CHURN PREDICTION; SELECTION; CHALLENGES; RETENTION;

D O I：

暂无

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Despite superior generalization performance Support vector machines (SVMs) generate black bar models. The process of converting such opaque models into transparent model is often regarded as rule extraction. This paper presents a new approach for rule extraction from SVMs using modified active learning based approach (mALBA), to predict churn in bank credit cards. The dataset is obtained from Business Intelligence Cup 2004, which is highly unbalanced with 93% loyal and 7% churned customers' data. Since identifying churner is paramount from business perspective, therefore considering sensitivity alone, the empirical results suggest that the proposed rule extraction approach using mALBA yielded the best sensitivity compared to other classifiers.

引用

页码：461 / +

页数：4

共 49 条

[1]

Adriaans P., 1996, DATA MINING

[2] Customer relationships and the small software firm -: A framework for understanding challenges faced in marketing [J].