Using machine learning algorithms to detect frauds in telephone networks

Abstract

This paper presents an analysis of voice traffic in telephone networks, based on machine learning algorithms to detect frauds made by callers. Starting from the raw data set that includes information about the call date, destination number, duration and caller's number, in our approach we were able to identify fraudulent calls in early stages. For balance, the data set was split in 2 parts: one for training and one for testing. To obtain mean’s values from dataset, a standardization technique was applied in order to scale the data before the dimensionality reduction using Principal Component Analysis. Then, the first two components were used as inputs for Logistic Regression and Random Forest models, having the caller as target. Finally, the target was moved on the destination file so as to identify the caller and the moment when the call has started based on a vector representation of words.