Comparative Evaluation of GPS Data Smoothing Techniques for Error Minimization in Sport Performance Metrics

This study aimed to refine GPS speed calculations by enhancing the precision of 10 Hz speed data from smart shin-mounted devices to closely approximate the 18 Hz data from a top-tier GPS vest. The research tested multiple combinations of merging and filtering techniques on the GPS speed data obtained from the right and left devices of twenty-five male football players from an Italian Serie D team during controlled and ecological sessions. The results revealed significant advancements in GPS speed data accuracy. The optimal strategy decreased the global RMSE from 2.24 km/h to 1.54 km/h and increased the global R² from 0.71 to 0.87, reflecting superior predictive accuracy. Despite a less aggressive filtering approach compared to the original method, the RMSSD value was maintained at 0.33. Additionally, the arithmetic mean RMSE across speed zones improved from 2.43 km/h to 1.88 km/h, and the RMSE in the number of sprints saw a notable reduction from 23.98 to 12.89. Although the global RMSE for max speed only modestly improved from 2.63 km/h to 2.47 km/h, this highlighted the strategies' emphasis on overall signal enhancement rather than peak event alteration. The study's findings indicate that the best strategy significantly improved the accuracy of GPS speed calculations compared to the original method, enhancing overall data reliability and reducing errors in speed measurement. This research paves the way for further analyses, particularly in improving acceleration signal accuracy and developing metabolic power metrics, which are invaluable in sports science and for the support staff of professional teams. Future studies could explore broader applications by including different athlete populations and pitch types to enhance the generalizability of the results.