This study explores sentiment analysis for Tigrigna social media texts, a low-resource language with limited computational tools. We collected and annotated a dataset of Tigrigna posts and comments from social media platforms and evaluated several machine learning and deep learning models, including Naive Bayes, SVM, LSTM, and XLM-RoBERTa. Results show that XLM-RoBERTa, a transformer-based model, achieved the highest performance with an F1-score of 0.83, effectively handling Tigrigna’s complex morphology and cultural nuances. Key challenges included data scarcity, dialectal variations, and idiomatic expressions unique to Tigrigna. Future work suggests expanding Tigrigna resources, exploring optimized models for resource-constrained environments, and developing applications for real-world sentiment monitoring. This research contributes to advancing NLP for low-resource languages, promoting inclusivity in sentiment analysis.