Fake-New-Detection

Project structure

Fake-News-Detection/
├─ dataset                    # contain our fake and real news dataset
├─ api.py                     # Flask API that handles URL prediction
├─ interface.py               # Streamlit interface for user interaction
├─ logistic_regression.ipynb  # Training notebook of LR model
├─ model_svm.nb               # Training notebook of SVM model
├─ model_nb.ipynb             # Training notebook of Naive Bayes model
├─ best_svm_model.pkl         # Saved best SVM model
├─ best_logreg_model.pkl      # Saved best Logistic regression model
├─ best_nb_model.pkl          # Saved best Naive Bayes model
├─ README.md

This project is a fake news detection system that takes a URL of a news article, scrapes its content, preprocesses it, and predicts whether the news is REAL or FAKE using a best trained machine learning model (SVM).

It includes:

• A Flask API (app.py) that handles URL input, scrapes the article using newspaper, cleans the text, and returns predictions.
• A Streamlit interface (interface.py) to input URLs and display predictions with probabilities and an article snippet.

---

Features

• Scrapes article text from any accessible URL using the newspaper3k library.
• Text preprocessing: lowercasing, removing special characters, lemmatization, removing stopwords (from both SpaCy and NLTK).
• Uses a trained SVM model for classification.
• Returns pseudo-probabilities for REAL/FAKE when the model does not support predict_proba.
• Displays predictions via a simple web interface with Streamlit.

---

How to run

1. Navigate to api.py and run python api.py on terminal.
2. Then onto another terminal, run streamlit run interface.py to run the interface.

When on the interface:

• Prediction (REAL or FAKE)
• Probabilities for each class
• First 500 characters of the cleaned article