Statistical models outperform neural networks for Parkinson's disease progression tracking

Researchers compared statistical and neural mixed-effects models for tracking Parkinson's disease progression using voice data. The study found that traditional Generalized Additive Mixed Models (GAMMs) outperformed more flexible neural models in a small cohort setting, achieving lower prediction error. This suggests that interpretable statistical models are more suitable for small longitudinal studies, while neural approaches may require larger datasets to avoid overfitting. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Suggests interpretable statistical models may be more effective than neural networks for small, longitudinal medical datasets.

RANK_REASON Academic paper comparing statistical and neural modeling approaches.

Read on arXiv stat.ML →

• Parkinson's disease
• Ran Tong
• Oxford Parkinson's telemonitoring dataset
• Generalized Additive Mixed Models
• Neural Mixed Effects models
• Generalized Neural Network Mixed Models
• arXiv

• paper
• other