Motor and Sensory Function-based Ensemble Learning Classifiers for Detecting Cognitive Decline Subtypes

Mild cognitive impairment (MCI) is the transitional stage between cognitively healthy aging and dementia, characterized by subtle neurocognitive changes, and can be categorized into two variants: amnestic and non-amnestic. These subtypes have different progressions and incidences of dementia, making detection challenging. On the basis of evidence from previous studies that investigated the motor and sensory functions in different MCI subtypes, we hypothesized that motor and sensory function variables can be used to develop a cost-effective classification tool that can distinguish cognitive decline subtypes. Community-dwelling men (N = 117) over 65 participated in this study. We assessed the participants’ motor-sensory function and developed classification models for dividing them into cognitive decline subtypes. In this study, we developed Random Forest (RF), Gradient Boosting, XGBoost, and Histogram-based Gradient Boosting models. RF was the most effective, with an accuracy of 0.833, an area under the curve 95% confidence interval ranging from 0.965 to 0.978, a specificity ranging from 0.835 to 0.988, and a sensitivity ranging from 0.469 to 0.932. Both the sensitivity and specificity for detecting non-memory-related cognitive decline were high; therefore, the motor and sensory functionbased classification models developed in this study are anticipated to aid in the diagnosis of MCI subtypes, especially non-amnestic MCI.