Toward Scalable Learning with Non-Uniform Class and Cost Distributions: A Case Study in Credit Card Fraud Detection

Verylarge databases with skewedclass distributions and non-unlformcost per error are not uncommonin real-world data mining tasks. Wedevised a multi-classifier meta-learningapproachto address these three issues. Ourempirical results from a credit card fraud detection task indicate that the approachcan significantly reduceloss due to illegitimate transactions.

• Type: Conference
• Year: 1998
• Where: KDD
• Authors: Philip K. Chan, Salvatore J. Stolfo
• Keywords: Data Mining, Fraud Detection Task, KDD 1998, Multi-classifier Meta-learningapproachto Address, Real-world Data Mining