🛡️ Job Scam Detector

An NLP-based classifier that flags fraudulent job postings and recruiter messages, with an interpretable breakdown of why a message looks like a scam — not just a black-box label.

🎯 Why this project

Online job scams — fake "work from home" offers, upfront "registration fees," requests for bank details before any real interview — are common and costly. This project was born out of a real experience reporting a job scam, and turns that experience into a tool: paste in a job posting or recruiter message, and get back a scam probability plus the specific red flags that triggered it.

🧩 Problem Statement

Most people spot scams only after losing money or sharing sensitive information. This tool gives a fast, explainable second opinion before you engage — combining a machine learning classifier with transparent, rule-based red-flag detection so the output is trustworthy and actionable, not just a score.

⚙️ Approach

1. Curated dataset — hand-collected scam and legitimate job posting examples covering common Indian and international scam patterns (upfront fees, urgency language, requests for bank/Aadhar details, informal-only contact channels) vs. normal hiring language.
2. Feature engineering — beyond TF-IDF text vectors, extracts explicit red-flag signals: urgency score, payment-request score, sensitive-info-request score, informal-channel score, "no interview needed" score, exclamation/caps usage.
3. Model — Logistic Regression over combined TF-IDF + scaled engineered features, with class_weight="balanced" for the two classes.
4. Explainability layer — a separate rule-based explain_flags() function surfaces which red flags fired in plain English, alongside the model's probability score.
5. Interactive app — a small Flask UI to paste in any message and get an instant check.

📈 Results

Metric	Score
Accuracy (held-out split)	1.00
F1-score	1.00

⚠️ Known limitation: the dataset is small (~49 examples) and cleanly separable, so these numbers are optimistic. In testing, single weak signals (e.g. the word "urgent" in an otherwise normal posting) can still trigger false positives. This is a solid proof-of-concept architecture — the real next step is scaling up the dataset with more diverse, ambiguous, real-world examples (e.g. from public scam-report datasets) to properly stress-test and calibrate it.

🛠️ Tech Stack

Python scikit-learn pandas Flask TF-IDF Logistic Regression

🚀 How to Run

git clone https://github.com/naax005/job-scam-detector.git
cd job-scam-detector
pip install -r requirements.txt

# Train the model
python src/train.py

# Run the web app
python app.py
# then visit http://127.0.0.1:5000

Or use it directly in Python:

from src.predict import predict

result = predict("Congratulations! Pay a $49 registration fee via Western Union to activate your work from home job today!!!")
print(result["label"], result["scam_probability"])
print(result["red_flags"])

📁 Project Structure

job-scam-detector/
├── data/
│   └── job_postings.csv          # curated scam vs legit examples
├── src/
│   ├── feature_engineering.py    # red-flag feature extraction + explanations
│   ├── train.py                  # trains TF-IDF + Logistic Regression model
│   └── predict.py                # scores new text, returns probability + flags
├── models/                       # saved model, vectorizer, scaler (generated)
├── templates/
│   └── index.html                # Flask UI
├── app.py                        # Flask web app
├── requirements.txt
└── README.md

🔮 Future Improvements

• Scale up the dataset using public scam-report datasets and real scraped job postings for better generalization
• Add a browser extension so people can check messages without leaving LinkedIn/WhatsApp
• Try a transformer-based model (DistilBERT) for better handling of subtle/ambiguous phrasing
• Add domain/URL reputation checks for links included in messages
• Track false positive rate specifically on legitimate postings that use urgent-sounding but normal hiring language

👤 Author

Alinas Ferdavus — LinkedIn

Built after personally encountering and reporting an online job scam via cybercrime.gov.in — this project is part of a broader interest in applying data science to cybersecurity and consumer protection.