README.md

metadata

title: Multi-Label Emotion Classification (DeBERTa Ensemble)
emoji: 🧠
colorFrom: indigo
colorTo: gray
sdk: gradio
sdk_version: 4.8.0
app_file: app.py
pinned: true

Multi-Label Emotion Classification using Deep Learning

This Space hosts the final solution for the Multi-Label Emotion Classification challenge.

The model is an Ensemble of two Fine-Tuned DeBERTa-v3-Base architectures, optimized for Macro F1-Score. Performance was enhanced using Weighted Loss and Per-Label Threshold Optimization, achieving a final Macro F1-Score of 0.845.

Project Details:

• Target Emotions: Anger, Fear, Joy, Sadness, Surprise (Multi-Label)
• Model Architecture: Ensemble of Fine-Tuned DeBERTa-v3-Base
• Key Techniques: Weighted Loss, Per-Label Threshold Optimization, TAPT.

---

Author: 22f3001086 - Vaishnavi Bhan