Back to Projects
Machine Learning

Fake News Detection System

A comprehensive fake news detection framework comparing traditional machine learning approaches with modern deep learning models. The project provides detailed performance analysis across multiple datasets and model architectures.

Tech Stack

7 technologies used

PythonScikit-learnTensorFlowKerasBERTHugging FaceNLP

Key Features

  • Traditional ML models: Naive Bayes, Random Forest, Logistic Regression, Passive Aggressive
  • Deep learning models: CNN and LSTM for text classification
  • Fine-tuned BERT using Hugging Face's Trainer API
  • TF-IDF and CountVectorizer feature extraction pipelines
  • Evaluation across multiple datasets (BuzzFeed, LIAR, Kaggle, Reuters)
  • Reproducible training and evaluation pipelines

Challenges Solved

  • Handling class imbalance across different datasets
  • Comparing models fairly given different computational requirements
  • Generalizing across news from different domains and time periods

Outcomes & Impact

  • Comprehensive benchmark of fake news detection approaches
  • Insights into trade-offs between model complexity and performance
  • Reusable pipeline for text classification experiments