In recent days, social media users are drastically increasing, and they are very interested in participating in discussions and expressing their feelings in the form of comments. Most of the users use their native language, which is written in English(Code-Mixed Language). But the existing sentiment classification models can analyze the text sentiment if it is in English vocabulary or the script is in the native language. If the YouTube comments are in the code-mixed language, existing methodologies’ performance is not promising. To solve this classification problem, we use the Paraphrase XLM-RoBERTa model. We train the model on Tamil, Malayalam, and Kannada code-Mixed language datasets, and achieve F1-scores of 71.1, 75.3, and 62.5 respectively. Our team ranks first, second and third on Tamil, Malayalam, and Kannada code-Mixed language datasets.