Mapana-Journal of Sciences - Fake News Detection in Low Resource Language Using Machine Learning Techniques and SMOTE

Aug 5, 2026

In the digital information era, fake content spreads rapidly across images, audio, videos, and text, undermining reliability and trust, especially in low-resource languages such as Tamil. This research evaluates traditional machine learning approaches for detecting and classifying fake versus real news where labeled data and advanced NLP tools are limited. Comparative experiments measure performance using F1 score across models including logistic regression, SVM, naive Bayes, KNN, decision trees, and random forests.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Abstract

Keywords

Introduction

Related Work

Methodology

Tamil News Dataset

Mapana – Journal of Sciences 2024, Vol. 23, No. 4, 121-136 ISSN 0975-3303|[https://doi.org/10.12723/mjs.71.7](https://doi.org/10.12723/mjs.71.7) Fake News Detection in Low Resource LanguageUsingMachineLearningTechniques and SMOTE Rajalakshmi Sivanaiah*, Angel Deborah Suseelan*, Sushanth Dilli Baskar* and Swathika Durairaj* Abstract the era of digital information. This paper discusses the critical issues in detecting fakecontent in news articles Tamil language, where the availability of labeled data and advanced natural language processing tools are limited. We employ traditional machine learning models to mitigate this problem, with particular emphasis on detecting and classifying fake and real content in the context of Tamil news. Our study explores the performance of different models like logistic regression (F1 score: 91%), support vector machines (SVM) (F1 score: 91%), naive Bayes (F1 score: 89%), k-nearest neighbors (KNN) (F1 score:70%), decision trees (F1 score: 91%), random forests score: 89%). By conducting a comprehensive comparative analysis of these models within the challenging linguistic environment of Tamil, we aim to provide insights into their suitability for detecting fake content in low-resource languages and draw meaningful comparisons between their performance. * Department of Computer Science and Engineering, Sri Sivasubramaniya [rajalakshmis@ssn.edu.in](rajalakshmis@ssn.edu.in), [angeldeborahs@ssn.edu.in](angeldeborahs@ssn.edu.in), sushanth2110209@ [ssn.edu.in](ssn.edu.in), [swathika2110791@ssn.edu.in](swathika2110791@ssn.edu.in) Mapana-Journal of Sciences, Vol. 23, No.4 ISSN 0975-3303 Keywords: 1. Introduction In an age characterized by the relentless surge of digital information, the rampant dissemination of fake content in images, audio, videos, and text has evolved as a formidable challenge, posing severe threats insidious, penetrating even the linguistic boundaries of low-resource languages, where resources and tools for effective detection are markedly scarce. This research paper delves into this pressing concern, in the context of less resource languages, with a particular focus on the Tamil language. linguistic re-sources and their underdeveloped Natural Language the realm of fake news detection. These languages, in stark contrast to their high-resource counterparts like English, often struggle to access fake news effectively. It is in this challenging linguistic landscape that very old language that contains a rich heritage and culture and serves as a compelling focal point for our study. Boasting over 70 million speakers worldwide, Tamil holds immense cultural and historical language in the context of NLP and fake content detection. This research paper seeks to bridge this gap by assessing the need learning models to achieve this, like k-nearest neighbors (KNN), logistic regression, naive Bayes, support vector machines (SVM), decision trees, and random forests, which are well-established and these models is evaluated using the F1 score, a metric that balances Sivanaiah et al. Fake News Detection in Low Resource Language precision and recall, providing a robust measure of their effectiveness The central objective of our research is to conduct a comprehensive comparative analysis of these models to determine their suitability for detecting fake news in low-resource languages. By doing so, we aim to contribute valuable insights into the development of effectual strategies for combating misinformation in linguistic environments with limited resources. We have discussed the related work, methodologies used, results, and discussions arising from our analysis of the machine learning models in the context of low-resource languages in the following sections. The ultimate goal is to advance our collective understanding of fake news detection and mitigation in less resource languages. 2. Related Work In the digital realm, the scarcity of the tools used

FAQs

Why is fake news detection difficult in low-resource languages like Tamil?

Low-resource languages often lack labeled datasets and advanced NLP tools, which reduces the availability of effective detection methods and increases vulnerability to disinformation.

Which machine learning models are evaluated for fake news detection?

The study evaluates logistic regression, support vector machines (SVM), naive Bayes, k-nearest neighbors (KNN), decision trees, and random forests.

How is model performance assessed in this research?

Performance is compared using the F1 score, which balances precision and recall to measure effectiveness for distinguishing fake and real news.

Aug 5, 2026

Upload by Evangeline