Table 4 Confusion matrix for fine-tuned GPT-4o-mini model (in %).

From: Comparing traditional natural language processing and large language models for mental health status classification: a multi-model evaluation

 

Depression

Normal

Suicidal

Bipolar

Stress

Anxiety

Personality disorder

Depression

88.33%

0.31%

10.81%

0.10%

0.10%

0.14%

0.10%

Normal

0.19%

98.55%

0.19%

0.00%

0.84%

0.19%

0.00%

Suicidal

19.76%

0.25%

79.85%

0.00%

0.05%

0.00%

0.00%

Bipolar

4.33%

0.19%

0.19%

93.03%

0.75%

0.56%

0.94%

Stress

1.42%

4.26%

0.20%

0.00%

92.10%

1.42%

0.61%

Anxiety

1.80%

1.25%

0.28%

1.11%

1.94%

92.63%

0.83%

Personality disorder

3.02%

0.50%

0.50%

0.50%

0.00%

1.01%

94.47%