News Authenticity Detection

Live Demo

Dataset

---

Project Overview

This project classifies news articles as Fake or Real using Machine Learning and Deep Learning approaches.

The pipeline uses TF-IDF features for classical models and LSTM embeddings for deep learning to capture context and improve accuracy.

Detecting fake news is crucial in today’s digital era to prevent misinformation from spreading.

---

Dataset

• Source: Fake and Real News Dataset - Kaggle

• Process:

  1. Two separate datasets: one for Fake news and one for Real news
  2. Combined both into a single dataset
  3. Cleaned the combined dataset (lowercasing, removing punctuation/special characters, stopwords, and optional lemmatization)

• Columns:

  Column	Description
  title	Headline of the news article
  text	Full content of the article
  subject	Category/subject of the news
  date	Date of publication
  label	0 → Fake, 1 → Real

---

Project Workflow

1. Data Cleaning

• Combined Fake and Real news datasets into one
• Lowercased all text
• Removed punctuation, special characters, and stopwords
• Optional lemmatization

2. Exploratory Data Analysis (EDA)

• WordClouds for Fake vs Real news to visualize frequent words
• Checked class distribution

3. Feature Engineering

• Combined title + text_clean for richer feature representation
• Extracted TF-IDF features (top 5000 unigrams and bigrams)

4. Classical Machine Learning Models

Model	Accuracy	Precision	Recall	F1-Score
Logistic Regression	~0.99	0.99	0.99	0.99
Random Forest	~0.998	1.0	1.0	1.0

5. Deep Learning: LSTM with Embeddings

• Tokenized and padded sequences
• Model architecture: Embedding → LSTM → Dense → Sigmoid
• Validation Accuracy after 4 epochs: ~0.992

Epoch	Train Accuracy	Validation Accuracy
1	0.9086	0.9858
2	0.9861	0.9850
3	0.9886	0.9878
4	0.9900	0.9916

---