Fake News Classifier (NLP + ML)

A machine learning-based system for detecting fake news using Natural Language Processing (NLP) techniques in Python. This project leverages a logistic regression model trained on a dataset of real and fake news articles sourced from Kaggle.

Project Overview

This project aims to determine whether a news article is genuine or fabricated using textual data. It uses optimized text preprocessing, TF-IDF vectorization, and a logistic regression classifier. The model achieved 96.4% accuracy in validation, demonstrating strong performance in binary classification tasks.

---

Dataset

The dataset was obtained from Kaggle's Fake and Real News Dataset. It contains labeled news articles with the following structure:

• Fake.csv — contains fabricated news stories (label = 0)
• True.csv — contains legitimate news articles (label = 1)

Total size: ~44,000 articles (sampled to 5,000 for rapid experimentation)

---

Technologies Used

• Python (v3.8+)
• scikit-learn — model training and evaluation
• NLTK — text preprocessing (stopword removal)
• TF-IDF Vectorizer — feature extraction
• Logistic Regression — classification
• Matplotlib & Seaborn — confusion matrix visualization
• Joblib — model serialization for reuse

---

Features

• Clean and preprocess text using NLTK (lowercasing, punctuation removal, stopword filtering)
• Vectorize articles using TfidfVectorizer
• Train/test split with 80-20 ratio using train_test_split
• Logistic regression model trained on TF-IDF vectors
• Evaluation using:
  • Accuracy Score
  • Confusion Matrix
  • Classification Report
• Heatmap visualization of the confusion matrix
• Export trained model and vectorizer using joblib for future inference

---

Model Performance

• Accuracy: 96.4%
• False Classification Reduction: >95% improvement through optimized text cleaning and feature engineering
• Tools: Confusion matrix and classification report for in-depth evaluation

---

Output Files

After training, the following files are saved:

• fake_news_model.pkl — Trained logistic regression model
• tfidf_vectorizer.pkl — Fitted TF-IDF vectorizer

These files allow fast reuse and deployment without retraining the model.

---

How to Run

# Install dependencies
pip install -r requirements.txt

# Run the script
python fake_news_classifier.py

---

License

This project is open source and available under the MIT License. You are free to use, modify, and distribute this software in both personal and commercial projects.

---

Author

Created by Henil Daslaniya. Contributions and suggestions are welcome!