Reliability and validity of the fulltrack AI application to determine cricket bowling line and length compared to 3D motion capture.

Abstract

This study examined reliability and validity of the Fulltrack AI application to identify cricket ball landing position (line, length). Nine hundred and thirty-two deliveries were compared to 3D motion capture, the criterion measure, with 836 included in analysis (516 bowled (pace = 420, spin = 96), 320 Sidearm^TM; 301 facing a batter). Agreement analysis indicated an intraclass correlation coefficient of >0.96 for raw and filter 3D line and length data, compared to Fulltrack AI. The coefficient of variation was acceptable for length (<10%) and larger for line (23.82%), albeit with a smaller standard error of measurement (SEM = 0.05 m), improving with outliers removed. Bland-Altman plots confirmed good statistical agreement between devices, with limits of agreement largely within maximal allowable difference values. There are potential practical application considerations, given SEM = 0.47 m for length (diameter of seven cricket balls); with greater variability detecting length closer to the batters-end, and line closer to the bowlers-end. Validity, using a generalised additive model, showed no significant differences between devices (p > 0.05), with no condition-based interaction effects. The Fulltrack AI application enables ecologically valid assessment of bowling performance. Considering the trade-off between this and the accuracy of information is warranted when deciding how best to apply it to coaching environments to support augmented feedback.