Detection of Telephony Spam and Scams using Recurrent Neural Network (RNN) Algorithm

Abstract

Malicious phone calls, including spam and scams, caused millions of global financial losses every year and was a difficult problem over many years. This work introduces the solution based on machine learning for telecommunications without underlying the telephone network infrastructure. The major obstacle of this ten years old problem is building efficient functions without access to the telephony network infrastructures. The previous Spam Call data set is collected first. The dataset includes several labelbased features to predict malicious calling. We primarily focus on using Recurrent Neural Network (RNN) algorithm to detect the malicious calls. With the proposed features, we review different state of the art methods of machine-learning, and it is inferred that the most optimal approach can minimize malicious calls to 90% while keeping over 90% of the binary call accuracy. The outcomes also show that without significant overhead latency the models can be implemented effectively with the help of an evaluation analysis.