Table 3 Comparison of the top-performing models across the three approaches for predicting the neuropsychiatric state score

From: Detecting neuropsychiatric fluctuations in Parkinson’s Disease using patients’ own words: the potential of large language models

Neuropsychiatric State Score Prediction Approach

Model

RMSE

MAE ± STD (Median AE, Q1-Q3)

R2

Semantic search

multi-qa-mpnet-base-dot-v1

14.0

11.7 ± 7.9 (11.0, 5.3–17.0)

0.43

Machine learning

gte-Qwen2-1.5B-instruct + Random Forests

11.2

8.7 ± 7.1 (6.2, 3.1–13.4)

0.64

LLM

Gemma-2 (9B) (few-shot setting)

10.6

8.1 ± 6.9 (6.0, 3.0–10.4)

0.68

  1. MAE mean absolute error, AE absolute error, Q1 first quartile, Q3 third quartile, RMSE root mean squared error, STD standard deviation, LLM Large Language Model.
  2. Best results are highlighted in bold.
Back to article page