An Intelligent System for Identifying Fraud Phone Calls Using Machine Learning Algorithms

Abstract

Phone fraud, also known as spam and unwanted calls, is a major problem in the telecom industry, costing millions of dollars annually all around the globe even as technology goes further and further ahead. The rise of phone scams, also known as spam using a mobile phone or a telephone, has become a common security risk for businesses and individuals alike. Machine learning and artificial intelligence have shown promising results in analyzing and detecting fraudulent or harmful phone calls. This paper gives an efficient method to forecast fraud calls through Call Detail Records (CDR) based on a selection of different machine learning and clustering algorithms. Data cleaning, normalization, feature selection, and data balancing based on SMOTE were deemed applicable to the CDR dataset that includes fundamental features, including caller ID, called ID, duration, cost, destination, and call type. The various models that have been used include DB scan, SVM, GCN, and XGBoost which revealed patterns of fraudulent behavior. The proposed XGBoost model had the best accuracy score of 96.7 %, illustrating that it could better recognize fraud than the others.