Optimizing Machine Learning Models for Accessible Early Cognitive Impairment Prediction: A Novel Cost-effective Model...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Public Health

Background:

• Cognitive impairment and dementia develop years before clinical manifestation.
• Undiagnosed dementia cases are prevalent, particularly in low- and middle-income countries, due to limited access to diagnostic tools.
• Accessible tools for early cognitive impairment diagnosis and prediction are lacking in scientific literature.

Purpose of the Study:

• To develop a cost-effective and accessible machine learning model for predicting cognitive impairment risk up to five years before clinical symptoms.
• To identify high-performing, computationally efficient models for early cognitive impairment detection.

Main Methods:

• Utilized National Alzheimer's Coordinating Center (NACC) Uniform Data Set (UDS) data for model training and evaluation.
• Developed a novel algorithm for selecting cost-effective, high-performance machine learning models.
• Performed feature selection, time-series analyses, and external validation of the selected model.

Main Results:

• The Support Vector Machine (SVM) model demonstrated superior cost-efficiency and performance compared to neural network models.
• Achieved an F2-score of 0.828 in cross-validation and 0.750 in a generalizability test.
• Demographic and historical health data were identified as crucial predictors for early cognitive impairment detection.

Conclusions:

• Machine learning offers a viable pathway for developing accessible and accurate tools for early cognitive impairment prediction.
• The developed SVM model provides a cost-effective solution for early risk assessment.
• Future efforts should focus on creating affordable assessment tools to support global dementia action plans.