Detecting SPIT Attacks in VoIP Networks Using Convolutional Autoencoders: A Deep Learning Approach

Voice over Internet Protocol (VoIP) is a technology that enables voice communication to be transmitted over the Internet, transforming communication in both personal and business contexts by offering several benefits such as cost savings and integration with other communication systems. However, VoIP attacks are a growing concern for organizations that rely on this technology for communication. Spam over Internet Telephony (SPIT) is a type of VoIP attack that involves unwanted calls or messages, which can be both annoying and pose security risks to users. Detecting SPIT can be challenging since it is often delivered from anonymous VoIP accounts or spoofed phone numbers. This paper suggests an anomaly detection model that utilizes a deep convolutional autoencoder to identify SPIT attacks. The model is trained on a dataset of normal traffic and then encodes new traffic into a lower-dimensional latent representation. If the network traffic varies significantly from the encoded normal traffic, the model flags it as anomalous. Additionally, the model was tested on two datasets and achieved F1 scores of 99.32% and 99.56%. Furthermore, the proposed model was compared to several traditional anomaly detection approaches and it outperformed them on both datasets.