Searching for Gait Markers of Cognitive-Motor Dual-task Interference using Machine Learning

Performing two concurrent tasks, or dual-tasking (DT), commonly results in reduced performance especially when compared to single tasking (ST). Two kinds of DT have been studied in largely separate literatures: concurrent cognitive tasks (cognitive-cognitive; CC) and concurrent cognitive task and motor tasks (cognitive-motor; CM). The present project’s long-term goal is to serve as a proof-of-principal for future studies to relate CC and CM DT in at-risk populations for high DT costs and consequences (e.g. falls in the elderly). The project’s immediate goal is to develop a more sensitive measure for CM DT. Healthy young adult participants (N=9) walked at a set pace, performed a demanding cognitive task (serial sevens subtraction), or performed both tasks concurrently. Motor behavior was tracked using a pressure-sensitive treadmill and 3D motion tracking. These data were reconstructed using Vicon Nexus software and custom preprocessing with MATLAB, iteratively divided into training/testing sets, and submitted to a linear support vector machine learning classifier. In order to better understand what aspects of the data support the ability to classify DT versus ST, the data were analyzed with a focus on mean position and movement variability over time (standard deviation). Results indicate classification is dictated by the mean (position of participant on the treadmill) rather than the standard deviation (change in movement characteristics). Future analyses will include classification based on higher-order features of the data and analysis of support vector machine feature weights to identify informative/uninformative features, potentially allowing streamlining of future data collection.