ML Fake News Classifier

[!Python](https://python.org) [!Scikit-learn](https://scikit-learn.org) [!NLTK](https://www.nltk.org) [!Pandas](https://pandas.pydata.org)

Machine learning classifier for detecting fake news articles using Natural Language Processing (NLP). Trained on labeled datasets to distinguish fact-based reporting from misinformation, using feature extraction and text classification techniques.

• Text Classification: Binary classifier (real/fake news)
• NLP Pipeline: Tokenization, vectorization (TF-IDF), and feature engineering
• Multiple Algorithms: Logistic Regression, Naive Bayes, Random Forest support
• Model Evaluation: Precision, recall, F1-score, confusion matrix reporting
• Serialized Models: Save/load trained classifiers for inference
• Customizable Features: Adjustable stop words, n-grams, and vocabulary size
• Web Integration Ready: Easy inference on new articles

Language: Python 3.8+
ML Framework: Scikit-learn
NLP Tools: NLTK, TextBlob
Data Processing: Pandas, NumPy
Vectorization: TF-IDF (scikit-learn)
Evaluation: Cross-validation, classification metrics

• Python 3.8+
• Scikit-learn, Pandas, NumPy, NLTK

git clone <repo-url>
cd ML-Fake-news-classifier

pip install -r requirements.txt
# Or manually:
pip install scikit-learn pandas numpy nltk textblob matplotlib

# Download NLTK data
python -m nltk.downloader punkt stopwords wordnet

The model expects a CSV with columns:

• text: Article body
• label: 0 (real) or 1 (fake)

Example datasets:

• Kaggle Fake and Real News Dataset
• MediaEval Veracity Dataset

# Train model on labeled dataset
python train.py --dataset news.csv --model logistic_regression --output model.pkl

# Evaluate on test set
python evaluate.py --model model.pkl --test_data test.csv

# Cross-validation
python train.py --dataset news.csv --cv 5

from classifier import load_model, predict

model = load_model('fake_news_classifier.pkl')
result = predict(model, "Article text here...")
print(f"Prediction: {'FAKE' if result == 1 else 'REAL'} (confidence: {confidence:.2%})")

Preprocessing:

• Lowercase conversion
• Punctuation & special character removal
• Stop word removal
• Lemmatization

Vectorization: TF-IDF (max 5000 features, bigrams)

Algorithms:

• Logistic Regression (baseline)
• Multinomial Naive Bayes
• Random Forest (ensemble)

Performance (typical):

• Accuracy: 92–96%
• Precision: 90–95%
• Recall: 91–96%

Top indicators of fake news:

• Clickbait language patterns
• Absence of source attribution
• Emotional manipulation keywords
• Unusual punctuation/capitalization

• Dataset bias (training data source affects generalization)
• Context-dependent: satirical content may be misclassified
• Language-specific: English-only training
• Evolving tactics: New misinformation patterns require model retraining

[Add your license]