Rule Extraction from Support Vector Machine Using Modified Active Learning Based Approach: An Application to CRM
| Autor: | S. Bapi Raju, Vadlamani Ravi, M.A.H. Farquad |
|---|---|
| Rok vydání: | 2010 |
| Předmět: |
Process (engineering)
business.industry Active learning (machine learning) Computer science Generalization Customer relationship management Machine learning computer.software_genre Support vector machine Black box Business intelligence Active learning Data mining Sensitivity (control systems) Artificial intelligence business computer |
| Zdroj: | Knowledge-Based and Intelligent Information and Engineering Systems ISBN: 9783642153860 KES (1) |
| DOI: | 10.1007/978-3-642-15387-7_50 |
| Popis: | Despite superior generalization performance Support vector machines (SVMs) generate black box 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. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|