---

Fake News Detection using NLP & Machine Learning

Project Overview

This project focuses on detecting Fake, Real, and Misleading news articles using Natural Language Processing (NLP) and classical Machine Learning techniques. The goal is to build a robust, explainable, and internship-ready system capable of analyzing news text and predicting its authenticity with confidence scores.

Key Features

• Multi-class classification: Fake | Misleading | Real
• TF-IDF based text vectorization (unigrams + bigrams)
• Passive Aggressive Classifier for fast and scalable learning
• Confidence-based inference for real-world usability
• Confusion Matrix & detailed classification metrics
• Clean, readable, and interview-ready code

---

Dataset

• Source: Public Fake News dataset

• Size: ~20,000 news articles

• Columns:

  • text → news content
  • label → fake / misleading / real

---

Machine Learning Pipeline

1. Data Loading & Inspection
2. Train–Test Split (Stratified)
3. Text Vectorization using TF-IDF
4. Model Training (Passive Aggressive Classifier)
5. Model Evaluation (Accuracy, Precision, Recall, F1)
6. Visualization using Confusion Matrix
7. Confidence-based Prediction on new samples

---

Model Performance

• Accuracy: ~90%
• Strong performance on Fake & Real classes
• Misleading class highlights real-world ambiguity in news classification

---

Sample Inference Output

Prediction: 🟥 Fake News | Confidence: 59.35%
Prediction: 🟨 Misleading News | Confidence: 57.38%
Prediction: 🟩 Real News | Confidence: 56.07%

---

Technologies Used

• Python
• Pandas, NumPy
• Scikit-learn
• Matplotlib, Seaborn
• NLP (TF-IDF)

---

Key Learnings

• How NLP converts text into numerical features
• Handling multi-class classification problems
• Importance of confidence scores in ML predictions
• Evaluating models beyond accuracy
• Understanding class imbalance and misleading content

---

Future Improvements

• Handle class imbalance using resampling
• Try transformer models (BERT)
• Improve misleading news detection
• Deploy as REST API using FastAPI

---

Author

Aryaman Jain Data Science Intern

---