Using Exploratory Data Analysis for Fraud Elicitation through Supervised Learning

Abstract

Outlier detection is a relevant problem for many domains, among which for detection of fraudulent behavior. Exploratory Data Analysis techniques are known to be very useful for highlighting patterns and deviations in data through visual representations. Less explored is the feasibility of using them to build learning models for outlier detection, which can be applied automatically to classify data without human intervention. In this paper we propose a method that uses one Exploratory Data Analysis technique -- Andrews curves -- in order to produce a classifier, which we applied to a real dataset, extracted from an online auction site, obtaining positive results regarding elicitation of fraudulent behavior.