Machine Learning-Based Classification of Multi-modal Fact-Checked Misinformation on Social Networks

¹ Department of Computer Science, HVPM COET, Amravati, (M.S.), India
² Department of Information Technology, HVPM COET, Amravati, (M.S.), India

Published online: 20 November 2025

Abstract

The rise of misinformation on social networks creates serious problems for public awareness, policy-making, and trust in society. Social media content is getting more complex, often including text, metadata, and multimedia. This makes it essential to have smart systems that can classify misinformation using various signals. This paper introduces a machine learning approach to check the misinformation that uses the MuMiN (Multilingual Multimodal Fact-Checked Misinformation) dataset. This dataset contains annotated claims, supporting evidence, user tweets, and fact-check labels. Structured preprocessing pipeline applied to get the dataset ready for analysis. The textual and structural features were extracted as features. Three machine learning models, Random Forest (RF), Gradient Boosting (GB), and a Stacking Classifier were developed and assessed. These models were evaluated using key performance metrics. The experimental findings indicate that the stacking ensemble regularly surpasses the individual base classifiers, attaining an accuracy rating of 89.12%. This highlights the advantages of combining models to manage complex, noisy, and multimodal social media data. This study emphasizes the value of merging multimodal feature representations with ensemble learning methods for effective and scalable misinformation detection on online platforms.