Adaptive-DecayRank: Real-Time Anomaly Detection in Dynamic Graphs

Overview

Adaptive-DecayRank is a novel anomaly detection framework for real-time anomaly detection in dynamic graphs. It leverages Bayesian PageRank updates with an adaptive decay factor to efficiently detect node anomalies and sudden structural changes.

Features

✔ Real-time anomaly detection on dynamic graph streams.
✔ Modified dynamic PageRank algorithm for improved detection accuracy.
✔ Efficient Adaptive Bayesian Updating for scalability.
✔ Outperforms AnomRank, SedanSpot, DynAnom, etc.

Dataset

We provide the following benchmark datasets for dynamic graph anomaly detection:

• DARPA – Cyber attack dataset Link

• CTU-13 – Botnet traffic dataset - CTU-13 dataset

• RTM-30 – Synthetic anomaly dataset

Dataset Preparation

1. Download & Unzip the datasets.
2. Move all extracted dataset files to the Adaptive-DecayRank folder.
3. Ensure the dataset files are in the same directory before compilation.

Compile the C++ code

-- run bash file ".\run.sh"

Summary of Code Structure

1. pagerank.cpp Implements Adaptive-DecayRank algorithm with Bayesian updating. Computes anomaly scores for each node in the evolving graph.
2. read_data.cpp Reads, processes, and prepares temporal graph data. Handles snapshot generation based on step size.
3. Algorithm Process The anomaly scores are stored in DecayRank.txt True Positive Rate (TPR) and False Positive Rate (FPR) are stored in tpr_fpr_forAUC.txt The AUC values are used to plot the Precision-Recall and AUC curve.

Installation

Clone the repository:

git clone https://github.com/YOUR_GITHUB_USERNAME/Adaptive-DecayRank.git
cd Adaptive-DecayRank