Performance Evaluation of Machine Learning based Robocalls Detection Models in Telephony Networks

Abstract

Many techniques have been proposed to detect and prevent spam over Internet telephony. Human spam calls can be detected more accurately with these techniques. However, robocalls, a type of voice spammer whose calling patterns are similar to those of legitimate users, cannot be detected as effectively. This paper proposes a model for robocall detection using a machine learning approach. Voice data recordings were collected and the relevant features for study were selected. The selected features were then used to formulate six (6) detection models. The formulated models were simulated and evaluated using some performance metrics to ascertain the model with the best performance. The C4.5 decision tree algorithm gave the best evaluation result with an accuracy of 99.15%, a sensitivity of 0.991%, a false alarm rate of 0.009%, and a precision of 0.992%. As a result, it was concluded that this approach can be used to detect and filter both machine-initiated and human-initiated spam calls.