Baloyi, MGRananga, SMbooi, Mahlatse S2026-01-192026-01-192025-05978-1-905824-74-8.http://hdl.handle.net/10204/14611This paper explores the effectiveness of models, specifically Multilingual Bidirectional Encoder Representations from Transformers (mBERT) and Random Forest, for detecting misinformation about Coronavirus Disease 2019 (COVID-19) in Xitsonga, a South African language. The focus is on evaluating how synonym replacement and translation techniques can enhance model performance for misinformation detection by comparing the mBERT and Random Forest models in terms of accuracy, precision, recall, and F1 score. The continuous spread of misinformation during the pandemic COVID-19 has highlighted the need for accurate and efficient classification systems, especially in low-resource languages. To address this challenge, we enhanced an English news dataset using two data augmentation techniques: synonym replacement and translation into Xitsonga using the Marian Machine Translation (MarianMT) model. We fine-tuned mBERT for sequence classification and compare its performance with Random Forest classification model using Term Frequency-Inverse Document Frequency (TF-IDF) features. The results demonstrate that mBERT achieves good accuracy, precision, recall, and F1 score compared to Random Forest. This paper contributes to the field of misinformation detection by introducing a multilingual approach and improving classification performance in low-resource languages.FulltextenCOVID-19Machine learningMisinformationDisinformationXitsongaConference Presentationn/a