A Data Mining System for Potential Customers Based on One-Class Support Vector Machine

Abstract

Abstract Commodity purchase data is usually severely skewed, which is reflected in the fact that there are far more negative data than positive data. This phenomenon makes it difficult for the binary classification model to obtain satisfactory results. Hence, we transform the binary classification problem into a one-class novelty detection problem. Specifically, this work proposes a potential customer mining system based on the One-class Support Vector Machine (OCSVM) and demonstrates its effectiveness for classification, prediction, and potential customer mining. This system allows merchants to focus on unpurchased customers with the strongest purchase intentions and to change their purchase decisions with minimal sale costs, which enables merchants to maximize their benefits.