🛡️ JobShield AI — Fake Job Detection System

An AI-powered job screening application that uses machine learning and text analysis to identify potentially fraudulent job postings.

📌 Project Overview

Online job scams can expose job seekers to financial fraud and misuse of personal information. JobShield AI analyzes job posting text and predicts whether a listing is likely legitimate or potentially fraudulent.

The application combines a trained machine-learning model with a simple warning-sign checklist to help users assess suspicious job advertisements.

🖥️ Application Preview

✨ Features

• Machine-learning-based fake job classification
• TF-IDF text feature extraction
• Logistic Regression classification
• Fraud risk score and configurable decision threshold
• Warning indicators for suspicious phrases
• Interactive Streamlit web interface
• Practical job-safety recommendations

🧰 Technologies Used

• Python
• Pandas and NumPy
• Scikit-learn
• TF-IDF Vectorizer
• Logistic Regression
• Matplotlib
• Streamlit
• Joblib
• BeautifulSoup

⚙️ How It Works

1. Collect and preprocess job-posting data.
2. Clean and combine relevant text fields.
3. Convert text into numerical features using TF-IDF.
4. Train and evaluate a Logistic Regression model.
5. Predict the likelihood of fraud-related classification.
6. Display the result, estimated risk score, and potential warning signs.

📊 Model Evaluation

The current model achieved approximately 97.34% test accuracy.

On the held-out test set, the fraudulent class had approximately:

• Precision: 69%
• Recall: 83%
• F1-score: 75%

The dataset is imbalanced, so accuracy alone does not describe the model's performance. The model may miss some fraudulent postings and may incorrectly flag legitimate ones.

🚀 Run Locally

1. Clone the repository

git clone YOUR_GITHUB_REPOSITORY_URL
cd JobShield

2. Create and activate a virtual environment

Windows:

python -m venv venv
venv\Scripts\activate

3. Install dependencies

pip install streamlit scikit-learn pandas numpy matplotlib beautifulsoup4 joblib

4. Launch the application

python -m streamlit run app.py

Open the local URL displayed in your terminal.

📁 Project Structure

JobShield/
├── app.py
├── README.md
├── requirements.txt
├── .gitignore
├── models/
│   ├── jobshield_model.pkl
│   ├── tfidf_vectorizer.pkl
│   └── threshold.pkl
├── notebooks/
│   └── 01_data_exploration.ipynb
└── src/
    ├── data_preprocessing.py
    ├── explain.py
    ├── feature_engineering.py
    ├── predict.py
    └── train_model.py

⚠️ Limitations

• Predictions depend on the training dataset and learned text patterns.
• Risk scores are not calibrated probabilities.
• Warning indicators are basic keyword-based checks.
• A low score does not guarantee that a job is genuine.
• Users should independently verify employers and never pay suspicious recruitment fees or share sensitive credentials.

🎯 Future Improvements

• Improve model performance on fraudulent listings.
• Add more explainable model predictions.
• Evaluate additional classification algorithms.
• Add a job-posting history and downloadable reports.
• Deploy the application online.

👩‍💻 Project Purpose

Developed as a Python and Data Science mini project to explore natural language processing, supervised machine learning, model evaluation, and interactive application development.