Machine Learning Approach to Multiclass Sentiment Analysis in Dravidian Language Code-Mixed Text

In this study, we focused on sentiment analysis of code-mixed text in Dravidian languages, specifically Tamil-English and Tulu-English, using advanced natural language processing (NLP) techniques. We address the challenges posed by linguistic diversity, the unique nature of code-mixing, and the limited availability of annotated datasets with the application of the Logistic Regression machine learning model enhanced by feature engineering. Our preprocessing approach involves normalizing input text, handling class imbalances, and applying sub word tokenization to accommodate the agglutinative and mixed nature of Dravidian languages. The model was trained for a multi-class sentiment classification task, distinguishing between the classes. Our experiment achieved an F1 score of 0.4409 on Tamil-English and 0.4509 on Tulu-English, this demonstrates the model's ability and limit to process Dravidian scripts and capture sentiments in code-mixed data on a multiclass dataset. Our methodology approach can be extended and improved to tackle other low-resource language challenges in future work. The code and dataset used are available at https://github.com/ABITCONSULT/Sentiment-Analysis-in-Dravidian-Language Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i2.1280Dimensions.Open Alex.