sajjadhadi/Mental-Disorder-Detection-Qwen2.5-0.5B-v1

Mental-Disorder-Detection-Qwen2.5-0.5B

Disclaimer

This mental disorder detection model is designed for research and educational purposes only. It is not intended for use in real-world medical environments or to detect actual mental health conditions. For any medical concerns, always consult a qualified healthcare professional. The developers assume no responsibility for misuse in clinical or real-life scenarios.

This model is a fine-tuned version of Qwen/Qwen2.5-0.5B.

Model description

This Large Language Model (LLM), based on Qwen 2.5 with 0.5 billion parameters, is fine-tuned on a dataset of 48,000 records to classify five distinct mental health-related classes: depression, suicide, bipolar, anxiety, and expressing. The model leverages lightweight quantization techniques (4-bit) for efficient computation while maintaining accuracy. Though designed for experimental and research purposes, it demonstrates the potential to assist in identifying mental health patterns through text, offering valuable insights for educational and academic applications.

Metrics

Class	Top-1 Accuracy	Top-1 Count	Top-2 Accuracy	Top-2 Count
Overall	0.68	1656/2450	0.80	1971/2450
Depression	0.80	687/861	0.94	813/861
Suicide	0.38	175/466	0.57	266/466
Bipolar	0.78	274/351	0.82	287/351
Anxiety	0.69	286/413	0.76	313/413
Expressing	0.65	234/359	0.81	292/359

For more information about the usage and metrics you can check this Kaggle notebook.

This model is performing much better on the Suicide class.