Fake News Credibility Classifier

SVM Linear Model System to evaluate news article credibility whether it is Fake or Real, trained on the WELFake dataset.

---

Overview

This project uses a Linear SVM model trained on the WELFake dataset to classify news articles as real or fake. The system combines machine learning with rule-based legitimacy checks for improved accuracy on current news.

---

Model Performance (Evaluation Metrics)

The model was trained and evaluated on 20,000 articles with the following metrics:

Metric	Value
Accuracy	94.30%
Precision	93.96%
Recall	94.88%
F1-Score	94.42%

Metric Definitions

• Accuracy: Percentage of correctly classified articles (both real and fake)
• Precision: Of articles predicted as fake, how many were actually fake
• Recall: Of all actual fake articles, how many were correctly identified
• F1-Score: Harmonic mean of precision and recall

---

Features

• Linear SVM classification model
• TF-IDF feature extraction (10,000 features)
• Legitimacy checks for news agencies and journalistic language
• Real-time credibility scoring (0–100)
• Confidence percentage display
• Current event detection (2024–2025 topics)

---

Project Setup Guide

Follow these steps to run the project locally.

1. Clone the Repository

git clone https://github.com/ashvin2005/AI_ML_project.git
cd AI_ML_project

2. Install Dependencies

pip install -r requirements.txt

If streamlit is not installed properly:

pip install streamlit

3. Run the Application

streamlit run app.py

Open browser at:

http://localhost:8501

---

Project Structure

├── app.py
├── model.ipynb
├── requirements.txt
├── svm_model.joblib
├── tfidf_vectorizer.joblib
└── .gitignore

---

Dataset

• Source: WELFake Dataset
• Size: 72,000+ news articles
• Labels: Real (0) and Fake (1)
• Training Sample: 20,000 articles (Milestone-1)
• Full dataset (72K) planned for Milestone-2

---

How It Works

Text Preprocessing

• Lowercasing
• URL removal
• Punctuation removal
• Digit removal

Feature Extraction

• TF-IDF vectorization
• Unigrams and bigrams
• 10,000 maximum features

Classification

• Linear Support Vector Machine (SVM)

Legitimacy Checks (Rule-Based)

• News agency identifiers (Reuters, AP, etc.)
• Journalistic language (said, according to, reported)
• Current events (2024, 2025, Gaza, Ukraine, COVID, elections)

---

Usage Instructions

1. Enter news article text in the text area (minimum 30 characters)
2. Click "Analyze" button
3. View results:
   • Classification: Real News or Fake News
   • Credibility Score: 0–100
   • Confidence Percentage
   • Debug Info: Legitimacy and fake indicator counts

---

Model Details

Detail	Value
Algorithm	Linear Support Vector Machine (SVM)
Features	TF-IDF with 10,000 max features
Stop Words	English
Training Time	~2–3 minutes (20,000 articles)
Future Scope	Training on full 72K dataset (Milestone-2)

---

Limitations

• Model trained mainly on 2016–2018 data
• May not recognize very recent events or names
• Best suited for formal news articles
• Short texts (<30 characters) not supported

---

Documentation and References

Dataset Reference

• WELFake Dataset: https://www.kaggle.com/datasets/saurabhshahane/fake-news-classification

Library Documentation

• Scikit-learn: https://scikit-learn.org/stable/
• Streamlit: https://docs.streamlit.io/
• Pandas: https://pandas.pydata.org/docs/
• NLTK: https://www.nltk.org/