FNMD: An Evaluation of Machine Learning and Deep Learning Techniques for Fake News Detection

Overview

FNMD is a research study focused on evaluating the effectiveness of machine learning and deep learning techniques for detecting fake news in social media content. With the rapid spread of misinformation online, this work explores robust computational approaches to distinguish between genuine and misleading information.

The paper investigates a combination of feature engineering techniques, sentiment analysis, statistical and linguistic features, and deep learning models to build an effective fake news detection framework.

---

Problem Statement

The rise of fake news across social platforms has created serious challenges for information reliability, public trust, and decision-making. Fake news is often intentionally designed to mislead readers and manipulate perception, making automated detection an important and urgent research problem.

---

Approach

This work proposes a hybrid methodology that combines:

• 🧠 Sentiment-based Features
  Sentiment scores are extracted at sentence level to capture emotional polarity and writing tone.
• 📊 Statistical & Linguistic Features
  Text-based structural and linguistic properties are used to enhance classification accuracy.
• ⚙️ Feature Selection Techniques
  Multiple feature selection methods are evaluated to identify the most relevant feature subsets.
• 🤖 Machine Learning Models
  Several classical ML classifiers are tested to determine optimal performance.
• 🧬 Deep Learning Models
  Advanced neural architectures, including LSTM, are applied to capture sequential and contextual patterns in text.

---

Datasets Used

The evaluation is conducted on multiple benchmark datasets:

• Liar Dataset
• GossipCop Dataset
• PolitiFact Dataset

These datasets provide diverse and real-world examples of fake and genuine news content.

---

Key Findings

• Feature selection significantly improves classification performance
• Hybrid feature engineering enhances model robustness
• Deep learning models outperform many traditional approaches in capturing contextual dependencies
• LSTM achieves up to ~88% accuracy, demonstrating strong performance in fake news detection tasks

---

Contribution

This study provides a comprehensive comparison between traditional machine learning and deep learning approaches for fake news detection. It highlights the importance of combining:

• Sentiment knowledge
• Linguistic features
• Feature selection techniques
• Deep sequence models

to build more accurate and reliable fake news detection systems.

---

Impact

FNMD contributes to the ongoing research in:

• Misinformation detection
• Natural language processing (NLP)
• Social media analytics
• Trustworthy AI systems