Fake news detection in Urdu language using machine learning.

Area of Science:

• Natural Language Processing
• Machine Learning
• Information Security

Background:

• The proliferation of fake news on social media necessitates effective detection methods.
• Existing fake news detection approaches often lack robustness for multi-domain datasets, particularly for Urdu news.
• Previous studies have utilized machine-translated datasets without manual verification, limiting real-world applicability.

Discussion:

• This research proposes a novel fake news classifier specifically designed for Urdu news.
• A diverse dataset of 4097 news articles across nine domains was collected and utilized.
• The study investigates the effectiveness of Term Frequency-Inverse Document Frequency (TF-IDF) and Bag of Words (BoW) with n-grams as features.

Key Insights:

• Feature stacking, combining text and verb features, significantly improves classifier performance.
• Ensemble models, including Random Forest (RF) and Extra Tree (ET), were employed for bagging and stacking.
• The proposed stacking model achieved high accuracy (93.39%), sensitivity (96.33%), and F1 score (93.17%), demonstrating robustness through fivefold and independent set testing.

Outlook:

• The developed classifier offers a promising solution for combating fake news in the Urdu language.
• Future research could explore advanced deep learning architectures for further performance enhancement.
• Validation on larger, more diverse, and real-world Urdu news datasets is recommended.