Leveraging Missing Values in Call Detail Record Using Naïve Bayes for Fraud Analysis

Abstract

In Telecom Fraud Management System (TFMS), the corrupted and missing values in call detail record (CDR) information are filtrated by rule-based classifier into the rejection proportion, as most of fraud detection approaches such as customer profiling and learning-based fraud detection technique were demanded clean information. Additionally, very few research works had discussed the issue of fraud detection in the area of corrupted CDR, due to its possibility yielded by lack of telecom equipment. In this paper, we present the leveraging missing values method, using the Naive Bayes approach posterior to rule-based classifier to analyze the probability of corrupted and missing values in CDR which then consequently led to the discovery of useable record lies in rejected CDR. This approach differs from other missing value handling and fraud analysis research where it does not attempt to treat the missing values; instead, leveraging it into the probabilistic model for fraud analysis. In this paper, we also reveal an impact of the missing values in telecommunication services and significantly vacant an avenues future work in fraud detection research.