Sensorimotor Activity Patterns using Machine Learning: Assessing the Impact of Auditory Timing Perception and Comparing Different Algorithms

Perception and action are the characteristics of the human condition and define our impact on the world around us. The brain's ability to effectively sort and incorporate these signals for flawless interactions with the environment gives rise to the hypothesis that these processes share an integrated set of representations, including our arbitrary sense of time. Interesting studies investigate how simultaneous activities affect our ability to keep track of time when we consciously move from one place to another. Surprisingly, no study has yet delved into the possibility of predicting simultaneous movements based on our time estimates or vice-versa. The purpose of this study is to examine the role of machine learning in the statistical prediction of sensorimotor tasks explicitly involving isolated auditory tones and arm motions, or both simultaneously based on conscious timed actions. The dataset was collected via Kaggle and consisted of 210 trials across 20 right-handed participants with each participant completing 14 trials in a block, for a total of 96 trials, and the sequence of the conditions was pseudo-randomized. Using data from seven distinct features representing a variety of consciously timed activities, nine ML regressor-type models have been trained to produce a statistical prediction of sensorimotor tasks. Each of the nine regressor models has its hyperparameters optimized using RandomizedSearchCV for better performance. The various performance matrices of ML algorithms were compared, including the MAE, RMSE, R
²
, MARE, NSE, and residual plots.