adicadi/medipal-mental-health

MediPal Mental Health Classification Model

Model Description

This is a fine-tuned RoBERTa-Large model for mental health sentiment classification. It classifies text into 7 categories:

• —Normal: Healthy mental state
• —Anxiety: Anxious thoughts and worry
• —Depression: Depressive symptoms
• —Stress: Stress-related content
• —Bipolar: Bipolar disorder indicators
• —Personality disorder: Personality disorder symptoms
• —Suicidal: Suicidal ideation (requires immediate intervention)

Model Performance

• —Accuracy: 76.5%
• —F1 Score: 76.6%
• —Base Model: RoBERTa-Large
• —Training Dataset: Mental health text corpus

Intended Use

This model is designed for mental health journal analysis and should be used as a supportive tool, not as a replacement for professional mental health diagnosis.

How to Use

Using Transformers

from transformers import RobertaTokenizer, RobertaForSequenceClassification import torch

model = RobertaForSequenceClassification.frompretrained("adicadi/medipal-mental-health") tokenizer = RobertaTokenizer.frompretrained("adicadi/medipal-mental-health")

text = "I feel anxious and worried" inputs = tokenizer(text, returntensors="pt", padding=True, truncation=True, maxlength=128)

with torch.nograd(): outputs = model(**inputs) predictions = torch.nn.functional.softmax(outputs.logits, dim=-1) predictedclass = torch.argmax(predictions).item()

labels = ['Anxiety', 'Bipolar', 'Depression', 'Normal', 'Personality disorder', 'Stress', 'Suicidal'] print(f"Predicted: {labels[predictedclass]}") print(f"Confidence: {predictions[predictedclass].item():.2f}")

Using Inference API

curl https://api-inference.huggingface.co/models/adicadi/medipal-mental-health -H "Authorization: Bearer YOURHFTOKEN" -H "Content-Type: application/json" -d '{"inputs": "I feel anxious and worried"}'

Limitations

• —Not a substitute for professional mental health diagnosis
• —May not capture cultural or linguistic nuances
• —Trained on English text only
• —Requires professional validation for clinical use

Ethical Considerations

This model deals with sensitive mental health data. Users should:

• —Handle predictions with care and empathy
• —Provide appropriate crisis resources for high-risk predictions
• —Not use as the sole basis for mental health decisions
• —Comply with local healthcare regulations

Contact

For questions or issues, please contact the model author.