smksean/nigeria-fake-news-detection

NLP classifier for detecting fake news in Nigerian media using machine learning and text feature extraction # Nigeria Fake News Detection > Binary NLP classifier detecting fake vs. real news from Nigerian media — TF-IDF pipeline with a deployed Streamlit web app covering 38 Nigerian news sources. --- ## Problem Misinformation spreads rapidly across Nigerian digital media. This project builds a classifier that flags fake news from 38 major Nigerian outlets — including Punch, Vanguard, Channels TV, Sahara Reporters, Arise News, and others — based on headline and article content. --- ## Approach 1. **Dataset**: Synthetic Nigerian news dataset with balanced real/fake labels across 38 sources 2. **EDA**: Label distribution, top news sources, headline length analysis by label, word clouds for fake vs. real news 3. **Preprocessing**: Regex cleaning, stopword removal, NLTK tokenisation 4. **Vectorisation**: TF-IDF 5. **Modelling**: Multiple scikit-learn classifiers evaluated; best model serialised to `models/model.pkl` 6. **Deployment**: Streamlit web app — paste a headline and select a news source to get an instant real/fake prediction --- ## Results - Trained model: `models/model.pkl` - Vectoriser: `vectorizer/vectorizer.pkl` - Full classification report and confusion matrix in the notebook --- ## How to Run **Install dependencies:** ```bash pip install -r requirements.txt ``` **Run the web app:** ```bash streamlit run app.py ``` **Explore the notebook:** ```bash jupyter notebook "notebooks/Nigeria_fake_new_detection (2).ipynb" ``` --- ## Tech Stack `Python` · `scikit-learn` · `NLTK` · `pandas` · `NumPy` · `Streamlit` · `matplotlib` · `seaborn` · `joblib` --- ## Project Structure ``` nigeria-fake-news-detection/ ├── app.py # Streamlit web app ├── models/model.pkl # Trained classifier ├── vectorizer/vectorizer.pkl # TF-IDF vectoriser ├── notebooks/ │ └── Nigeria_fake_new_detection (2).ipynb # EDA + training ├── data/ │ └── nigeria_synthetic_news_dataset.xlsx ├── output/ …