Development of an Evaluation Model for Cognitive Decline by Quantification of Motor Learning

Herein, we developed a discriminant model for cognitive decline by quantifying motor learning as a feature variable. Motor learning has been investigated from the contexts of both motor and cognitive domains. However, previous studies separately analyzed a motor learning results, which may have led to incorrect results. The result of each domain has a different measurement meaning, meaning that the motor learning results must be obtained separately. In this context, we quantified motor learning by separating motor and cognitive domains. This study aimed, the obtained motor learning results were used to build a discriminant model. However, when machine learning algorithms were applied, it was not clear which machine learning method was best depending on the feature variables. Therefore, in preliminary experiments, we compared the machine learning methods K-nearest neighbor (KNN), support vector machines, random forests, and AdaBoost methods, finding that KNN had the best accuracy. As such, we focused on KNN in this study. Our results revealed that at 10 trials, the accuracy was 0.57, at 30 trials accuracy was 0.65, and at 50 trials the accuracy was 0.96. However, measurement took about 30 minutes. These motor learning results suggest that 30 minutes of motor learning training is required to optimize the accuracy. Further, these motor learning results were shown to predict cognitive decline in a mild cognitive impairment (MCI).