Fake News Classifier - NLP Project

This repository contains a full Machine Learning pipeline to classify news as real (1) or fake (0) using Natural Language Processing (NLP) techniques, built entirely in Python.

The project includes preprocessing, feature engineering, model training with cross-validation, and prediction on unseen validation data.

---

Final Deliverables:

• predictions_validation.csv
• fake_news_nlp.ipynb

Project Structure

• dataset:
  • data.csv # Labeled dataset for training and validation
  • validation_data.csv # Unlabeled dataset to predict
• results:
  • predictions_validation.csv # Final predictions on validation data
  • predictions_validation_onlytext.csv # Final predictions on validation data with only text model
• models:
  • random_forest_model.pkl # Saved trained model
  • tfidf_vectorizer.pkl # Saved TF-IDF vectorizer
  • random_forest_model_onlytext.pkl # Saved trained model with only text as a feature
  • tfidf_vectorizer_onlytext.pkl # Saved TF-IDF vectorizer with only text as a feature
• fake_news_script.py # Main Python script
• fake_news_nlp.ipynb # Jupyter Notebook step by step how-to process to train the model
• only_text_nlp_model.ipynb # Jupyter Notebook training a new model using only text as feature
• README.md # Project description
• requirements.txt
• first_steps_data_exploration.ipynb # Jupyter Notebook with the first steps we took exploring the data

How to Run

1. Install the dependencies:

pip install -r requirements.txt
python fake_news_script.py 

The script will:

- Preprocess the data (tokenize, remove stopwords, apply stemming)
- Combine title + text into a single feature
- Apply TF-IDF vectorization + One-Hot Encoding for subject
- Train a RandomForestClassifier
- Perform 5-fold cross-validation with visual output
- Save the model and vectorizer
- Predict on validation_data.csv and save results in predictions_validation.csv
- Evaluation: Cross-Validation

Results:

We use 5-fold cross-validation to evaluate the model’s ability to generalize. - Mean accuracy: e.g. 99.8% - Standard deviation: e.g. ± 0.0013

Developed as part of the Ironhack Data Science program — NLP Challenge.

Authors: Affiong Akpanisong & Mo Benet