Detecting Abusive Comments in Mizo: A Machine Learning Approach for a Low-Resource Language

The detection of abusive language in online spaces is crucial for ensuring a safe digital environment, particularly for low-resource languages like Mizo. Mizo, a tonal Tibeto-Burman language spoken primarily in Mizoram, India, poses significant computational challenges due to its phonetic complexity and limited linguistic resources. This research presents a method based on machine learning for abusive comment detection in Mizo, addressing the lack of annotated datasets and specialized NLP tools. A structured pipeline involving data collection, preprocessing, feature engineering, and model evaluation was implemented. Our study compares the effectiveness of several conventional machine learning methods, such as Random Forest, Support Vector Machines (SVM), Logistic Regression, and XGBoost, against transformer-based models such as Multilingual BERT(mBERT) and MizBERT. According to experimental data, MizBERT achieves the highest accuracy and F1-score, outperforming all other models by a substantial margin. This work contributes to the development of computational tools for Mizo NLP, laying a foundation for automated moderation systems and fostering digital inclusivity for Mizo-Speaking communities.