Comparing Statistics and Machine Learning to Detect Insincere Grip Force Testing Using Manugraphy

Background Currently, there are no tests that have been proven to be capable of rating an individual's grip force measurement as sincere or insincere. However, different parameters have been found to vary in grip force testing for maximal versus submaximal effort. A novel data analysis and processing approach might be key to improving these measurements. This study explores the use of a machine learning (ML) algorithm as a means to more accurately determine the sincerity or insincerity of grip force testing. The ML algorithm compares the hand's load distribution pattern with the information generated using conventional statistical methods. Methodology This study uses manugraphy data collected as part of a previous investigation that analyzed load distribution patterns of the right and left hands of 54 healthy subjects. The subjects underwent grip force testing using maximal or submaximal effort, and the percentage contributions of each of the seven defined anatomical areas of the hand were calculated with respect to the total load applied. The predictions based on the load distribution and its use for rating individual grip force measurements as sincere or insincere were compared with the results of conventional statistical methods (thresholds for a bi-manual area-to-area comparison) and an ML algorithm. Results Based on an area-to-area comparison, our method achieved a sensitivity of 54% and a specificity of 78% to detect insincere effort. A predictive ML model developed using these data was capable of recognizing submaximal effort based on the hand's load distribution pattern, determining a sensitivity of 94% and a specificity of 99%. Conclusions Compared to conventional methods, the use of an ML algorithm considerably improved the validity of manugraphy results in discerning the sincerity or insincerity of grip effort.