Internship Scam Detection

A full-stack machine learning application designed to detect fraudulent or spam internship postings. This project uses a trained supervised learning model (Decision Tree) served via a Flask backend, with a responsive web interface for users to analyze job descriptions and requirements.

---

Live Demo: https://spam-detect-saj2.onrender.com

Currently uses Flask's development server. Production deployment with Gunicorn is planned.

---

spam-detect.walkthrough.mp4

---

Project Structure

Internship Scam Detection/
├── data/
│   └── internship_scam_data.csv       # Raw dataset used for training
├── models/
│   ├── model.pkl                      # Trained machine learning classifier
│   └── pipeline.pkl                   # Data preprocessing pipeline (ColumnTransformer/Imputer)
├── notebooks/
│   ├── internship_scam_detection.ipynb # Jupyter notebook for EDA and model training
│   └── README.md                      # Notes specific to the modeling process
├── static/
│   └── styles.css                     # Custom CSS for the web frontend
├── templates/
│   └── index.html                     # Main HTML interface
├── .gitignore                         # Specifies files for Git to ignore (e.g., venv, __pycache__)
├── app.py                             # Flask backend server and API routes
├── README.md                          # Project documentation
└── requirements.txt                   # Python dependencies (Flask, scikit-learn, pandas, etc.)

Installation & Setup

To run this project locally, follow these steps:

1. Clone the repository

git clone <your-repository-url>
cd "Internship Scam Detection"

2. Create a virtual environment A virtual environment keeps your project dependencies isolated.

python -m venv venv

3. Activate the virtual environment

• On macOS/Linux:

source venv/bin/activate

• On Windows:

venv\Scripts\activate

4. Install dependencies

pip install -r requirements.txt

Usage

Once your environment is activated and dependencies are installed, start the Flask server:

flask run

Open your web browser and navigate to http://127.0.0.1:5000 to view the application. Enter the details of an internship posting, and the ML pipeline will analyze the inputs and return a confidence score indicating the likelihood of it being a scam.

Model Methodology (notebooks/internship_scam_detection.ipynb)

The machine learning core of this application was built and optimized using scikit-learn in a Jupyter Notebook. The notebook walks through the entire lifecycle from raw data to a deployed model.

1. Data Preprocessing (The Pipeline) To handle incoming web data cleanly, a ColumnTransformer pipeline was built:

• Imputation: Missing numerical values (like stipend) are automatically filled with 0s using a SimpleImputer.
• Scaling & Encoding: Numerical data is normalized using StandardScaler, while categorical text (like work_mode or company_size) is transformed into a machine-readable format using OneHotEncoder.
• Feature Selection: High-cardinality fields (like company_name and location) were dropped to prevent the model from overfitting on highly unique text strings.

2. Model Training & Optimization

• Algorithm: The core engine is a DecisionTreeClassifier.
• Hyperparameter Tuning: Instead of guessing the best settings, GridSearchCV was used to systematically test multiple combinations of parameters (max_depth, min_samples_split, criterion) to find the most mathematically optimal tree.

3. Performance Metrics The final optimized model achieved the following results on the isolated test set:

• Overall Accuracy: 70%
• Scam Precision: 74% (When it flags a scam, it is correct 74% of the time)
• Scam Recall: 63% (It successfully catches 63% of all actual scams)

Technologies Used

• Frontend: HTML5, CSS3, Vanilla JavaScript (Fetch API)
• Backend: Python, Flask, Flask-CORS
• Machine Learning: scikit-learn, pandas, joblib

---

Created by

Naina Bhatnagar

• Socials
• Connect on linkedin
• Connect on github

---