This study uses Natural Language Processing (NLP) and machine learning techniques to explore lyrics-based music genre classification. By applying TF-IDF vectorization to a diverse, multilingual dataset of song lyrics, classification models such as Random Forest, Logistic Regression, and Linear SVC were trained to predict genres like rock, rap, and pop. The models achieved high accuracy in French, German, and Portuguese, while performance was lower in languages with fewer data samples, such as Norwegian. Rap and pop were classified well, but rock remained challenging due to lexical overlap in the dataset. This approach could improve music recommendation systems, content organization, and playlist generation. Future research could explore hybrid models, combining TF-IDF with transformer embeddings or incorporating phonetic patterns and sentiment analysis, to refine genre predictions, especially for low-resource languages.