SIP-Classifier

This repository contains the dataset and supporting scripts used in the paper "SIP-Classifier: Unsupervised Classification of SIP-IMS Signaling with Transformer and Clustering".

Repository Structure

• sip_dataset.csv: CSV dataset of SIP signaling flows. Columns:

  • Signaling Flow: a string describing the SIP signaling sequence.
  • Class: numeric label for the sequence (0 = benign, 1 = anomalous, 2 = unknown).

• sota/: folder containing baseline model implementations and results used for comparison in experiments.

  • CNN.py: Convolutional Neural Network baseline from the paper Detection of abnormal SIP signaling patterns: a deep learning comparison (Computers, 2022)
  • corr.py: Correlation-based baseline from the paper Correlation-Based Abnormal SIP Dialog Identification: A Performance Comparison with Bayesian and Deep Learning Approaches (IEEE Access, 2025).
  • HMM.py: Hidden Markov Model baseline from the paper A Machine Learning Approach for Prediction of Signaling SIP Dialogs (IEEE Access, 2021).
  • LSTM1.py and LSTM2.py: LSTM baseline implementations from the paper Classification of Abnormal Signaling SIP Dialogs Through Deep Learning (IEEE Access, 2021)
  • results/: directory containing the results obtained from running the baseline models on the sip_dataset.csv.

• util/: utility scripts and labeled subsets.

  • anomalous.csv: subset of the sip_dataset.csv containing only anomalous signaling flows, without duplicates.
  • benign.csv: subset of the sip_dataset.csv containing only benign signaling flows, without duplicates.
  • split.py: helper script to extract the anomalous.csv / benign.csv subsets.
  • stats.py: script to compute statistics over the dataset sip_dataset.csv.

Cite us

If you find this work interesting and use it in your academic research, please cite our paper!