What factors influence golf shot outcome: classical statistical analysis vs. heuristic possibilities
Brooks, Peter and Goss-Sampson, Mark (2010) What factors influence golf shot outcome: classical statistical analysis vs. heuristic possibilities. In: The British Association of Sport and Exercise Sciences (BASES) 2010 Annual Student Conference, 31 Mar - 1 Apr 2010, Aberystwyth, Wales. (Unpublished)Full text not available from this repository.
Kinematic research of the golf swing has focused on significant differences based on handicap with an assumed subject base homogeneity. Within these studies, invariably, shot outcome has not been quantified. The aims of this study were to analyse group homogeneity, then investigate whether kinematic differences exist based on shot quality, using classical statistical methods compared to Data Mining methodology (C5.0, PASW Modeller). Nine golfers (handicap: mean 1.4 s = 2.44) each performed 50 drives into a Kevlar screen. Informed consent was obtained in accordance with University ethics guidelines. Shot scores were calculated using a VectorPro launch monitor (Accusport, USA). Good/bad shots were designated as those ± 1 s from the mean. Kinematic data were captured at 500Hz (Qualisys, Sweden), 64 retro-reflective markers were tracked to define a 14 segment 6-DOF model (Visual 3D, C-motion, USA). Events were defined as initiation (INIT), start (ST), mid-backswing (MBS), top (TOP), mid-downswing hands (MDH), mid-downswing club (MDC) and contact (CON). Post data-cleaning, 436 shots remained for analysis. TOP-CON lateral CoG shift (mean -0.11m s = 0.037m) and shoulder rotation (mean 118º s = 15.25º) were selected for 2-step cluster analysis to assess group homogeneity. Four clusters were clearly defined and validated by k-means analysis. Good/bad shots within clusters were analysed using two-way ANOVA with Bonferroni adjustment. 18 variables associated with shoulder, torso and pelvic rotation at events, and 10 associated with weight transfer and stance were investigated. Significance was set at P < 0.05. Significant differences between shots were only found in Cluster1, these being pelvic-torso angle (CON) and shoulder-torso angle (MDC). Two clusters were further analysed using C5.0., the Neural Network ranks variables based on information gain and reducing entropy, discovering hidden patterns of information, (mean ± s), Cluster1- low rotation (93.87 ± 2.71º) with low lateral CoG shift (-0.077 ± 0.008m), Cluster3- high rotation (131.04 ± 2.39º) with high lateral CoG shift (-0.151 ± 0.021m). Three variables in Cluster1 were ranked in order of importance by the network, pelvic-torso angle and shoulder-torso angle (CON), shoulder-torso angle (ST). Although nothing significant was found using classical analysis in Cluster3, C5.0 discovered important variables related to shot outcome to be, weight distribution (ST), pelvic-LAB angle (MDC) with minor influences of pelvic-torso and shoulder-torso angles (MBS). Results suggest homogeneity of subject base should be analysed in future studies. Classical statistical analysis should be combined with heuristic analytical tools in sequential sporting motion to aid sports scientists and coaches in performance analysis.
|Item Type:||Conference or Conference Paper (Paper)|
|Additional Information:||This paper was part of invited oral presentation (stream 8), given on 31 March 2010, at the The British Association of Sport and Exercise Sciences (BASES) 2010 Annual Student Conference, held in Aberystwyth, Wales.|
|Uncontrolled Keywords:||golf, biomechanics|
|Subjects:||G Geography. Anthropology. Recreation > GV Recreation Leisure
R Medicine > RC Internal medicine > RC1200 Sports Medicine
|Faculty / Department / Research Groups:||Faculty of Engineering & Science > Department of Life & Sports Sciences|
|Last Modified:||05 Dec 2016 12:21|
Actions (login required)