Fake Account Detector

NLP-based machine learning system to classify fake social media accounts, developed as part of a KSCST-funded research initiative.

Problem

Fake accounts on social media platforms undermine trust, spread misinformation, and enable fraud. Manual detection doesn't scale to millions of accounts.

Approach

• Analyzed 10,000+ social media account records
• Extracted NLP-based features from user bio, post patterns, and account metadata
• Trained and benchmarked multiple classification models
• Validated results using precision, recall, and ROC-AUC to handle class imbalance

Results

Metric	Score
Accuracy	84%
Evaluation	Precision, Recall, ROC-AUC
Dataset size	10,000+ accounts

Tech Stack

Python · Scikit-learn · Pandas · NLP · REST APIs

Project Structure

fake-account-detector/
├── data/               # Dataset (fake_accounts.csv)
├── scripts/
│   ├── train.py        # Model training pipeline
│   └── predict.py      # Inference script
└── models/             # Saved model (fake_acc_model.pkl)

Setup

git clone https://github.com/5ushh/fake-account-detector
cd fake-account-detector
pip install pandas scikit-learn
python scripts/train.py

Background

This project was funded by KSCST (Karnataka State Council for Science and Technology) as part of undergraduate research at New Horizon College of Engineering (2023).