alfiyahqthz/sccmr-multiclass

SCCMR Multiclass: Depression/Anxiety/Stress Classification

Fine-tuned from SCCMR-MDA for multiclass mental health classification.

Model Description

Classifies text posts into one of three mental health conditions:

• Depression (Class 0)
• Anxiety (Class 1)
• Stress (Class 2)

Performance

Validation Results (Epoch 9):

• Accuracy: 0.8880
• Loss: 0.5826
• F1 (Macro): 0.8729

Architecture

• Base: Pre-trained SCCMR-MDA (trained on 21k DASS-21 posts)
• Text Encoder: BERT-base-uncased
• Projection: 768 → 512 (pre-trained)
• Classifier: 512 → 256 → 3

Usage

import torch
from transformers import BertTokenizer

# Load model and tokenizer
model = torch.load("pytorch_model.bin")
tokenizer = BertTokenizer.from_pretrained("alfiyahqthz/sccmr-multiclass")

# Predict
text = "I feel so depressed and worthless"
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
logits = model(inputs['input_ids'], inputs['attention_mask'])
prediction = logits.argmax(dim=1).item()

# prediction: 0=Depression, 1=Anxiety, 2=Stress

License

MIT