Fake Account Detection in Social Media

Using Graph Analytics + Machine Learning | BDA Mini Project

---

Problem Statement

Social media platforms are flooded with fake accounts (bots) that:

• Spread misinformation
• Manipulate trends
• Inflate engagement artificially

Manual detection is impossible at scale.

This project builds an automated detection system using:

• Graph Analytics (main approach)
• Machine Learning (enhancement)

---

Project Goal

Detect fake accounts using graph structure first, then improve accuracy using machine learning.

---

📁 Project Structure

fake_account_project/
│
├── data/
│   └── dataset.csv
│
├── generate_data.py
├── graph_build.py
├── feature_engineering.py
├── model.py
└── main.py

Dataset

• 5000 users (2500 bots, 2500 real)
• Synthetic but realistic behaviour

Bot Behaviour:

• High retweets & mentions
• Low followers
• New accounts

Real Users:

• Moderate activity
• Higher followers
• Older accounts

---

Graph Construction

• Each user = Node
• Interaction = Edge

Edges are created using:

• Retweet behaviour
• Mention behaviour

---

Graph Analytics

Metric	Meaning	Bot Pattern
PageRank	Importance of user	Low
In-degree	Incoming links	Low
Out-degree	Outgoing links	High
Clustering	Community connectivity	Low

---

Graph-Based Detection

graph_score = follower_retweet_ratio
+ pagerank
- mention_retweet_ratio

• Low score → Bot
• High score → Real user

Accuracy: ~87%

---

Community Analysis

• 1500+ communities detected
• Many isolated nodes

Bots are mostly isolated
Real users form connected clusters

---

Feature Engineering

• follower_retweet_ratio
• mention_retweet_ratio
• account_age_days
• hashtags presence

---

Machine Learning Models

• Random Forest
• Gradient Boosting
• Logistic Regression

---

Results

Method	Accuracy
Graph Detection	~87%
Random Forest	~98%
Gradient Boosting	~98%
Logistic Regression	~97%

Graph gives structure
ML improves accuracy

---

How to Run

pip install pandas networkx scikit-learn matplotlib

python generate_data.py
python main.py

---

Limitations

• Synthetic dataset
• Approximate graph
• No temporal behaviour analysis

---

Future Work

• Real Twitter API data
• NLP features
• Temporal analysis
• Real-time detection

---

🛠️ Tech Stack

• Python
• NetworkX
• scikit-learn
• pandas
• matplotlib

---