This study aims to analyze public sentiments towards Rohingya refugees in Indonesia using the IndoBERT model. We collected sentiment data from social media platforms and news articles, followed by preprocessing techniques including tokenization, cleaning, case folding, stemming, and filtering. Sentiment labels were assigned using the InSet lexicon, and the IndoBERT model was trained with these labeled data. Our findings reveal that the predominant sentiment is negative, with 65% of the sentiments classified as negative, 20% as neutral, and 15% as positive. The model demonstrated robust performance with an accuracy of 87%, precision of 85%, recall of 83%, and an F1 score of 84%. This research addresses a gap in sentiment analysis studies related to refugee issues and provides valuable insights into public perceptions, which could inform policies and interventions aimed at improving refugee integration and support systems in Indonesia.