This paper presents our system for the RANLP 2025 shared task on multilingual binary sentiment classification for Task-2 Spanish datasets for domains including social media and customer reviews. We experimented with various models from traditional machine learning approaches—Naive Bayes and LightGBM—to deep learning architectures like LSTM. Among them, the transformer-based XLM-RoBERTa model performed best with an F1 of 0.85, demonstrating its promise for multilingual sentiment work. Basic text preprocessing techniques were used for data quality assurance and improving model performance. Our comparison reflects the superiority of transformer-based models over the traditional methods in binary sentiment classification for multilingual and low-resource environments. This study enables the development of cross-lingual sentiment classification by establishing strong baselines and paying close attention to model performance in joint task settings.