Effective detection of Thai fake news using machine learning method

Abstract

The rapid spread of misinformation on digital platforms has become a critical challenge affecting public trust, social stability, and policy communication in Thailand. The aim of this study is to develop and evaluate a reproducible machine learning framework for Thai fake news detection from real world verified data from the Anti-Fake News Center Thailand during 2019 - 2024. The data set comprises four key data groups including governmental policies, health products, financial and stocks as well as disaster news. The suggested research methodology entails thorough data cleaning and text normalization, sentiment and emotions analysis through the usage of AI for Thai, dimension reduction by selecting relevant features and dealing with unbalanced classes through Synthetic Minority Oversampling Technique (SMOTE). For evaluation, three supervised learning algorithms (Naïve Bayes, Support Vector Machine [SVM] and neural networks) were tested using 80:20 and 90:10 train-test split ratios, K-Fold Cross-Validation and hyperparameter optimization. Based on the experimental findings, the neural network classifier achieved better accuracy rates, that is, 92.39%, with other performance metrics including precision, recall, F1-score, ROC-AUC and AUC-PR at 92.64%, 92.39%, 92.38%, 96.79%, 96.07% respectively. Therefore, the methodological framework was effective in identifying false news stories in the Thai language due to the use of linguistic features and oversampling strategies. The main contribution of this study is an interpretable baseline framework that integrates linguistic, affective, and metadata features for Thai fake news detection in a low-resource language setting.