ObakengTha/NLP---Bantu-Languages-Xhosa-

Multilabel emotion classification for isiXhosa using AfroXLMR on the BRIGHTER dataset. This repository contains data preprocessing, model fine-tuning, evaluation, threshold optimization, and explainability (SHAP & LIME) for low-resource African language emotion analysis. isiXhosa Emotion Classification with AfroXLMR: This repository presents a multilabel emotion classification model for isiXhosa using the BRIGHTER dataset. The project fine-tunes AfroXLMR to predict six emotions (anger, fear, joy, love, sadness, and surprise) and includes data preprocessing, class imbalance handling, threshold optimization, model evaluation, and explainability using SHAP and LIME. Features: Fine-tuned AfroXLMR for multilabel emotion classification BRIGHTER isiXhosa dataset Class imbalance handling with weighted loss and oversampling Threshold optimization for improved Macro-F1 Explainability with SHAP and LIME Evaluation using Micro-F1, Macro-F1, precision, recall, and per-label metrics