Simplified Swarm Optimization with Sorted Local Search for golf data classification

Abstract

Golf Swing is one of the most difficult techniques in sports to perfect, and a smooth swing can't be achieved without a correct process of bodyweight transfer between the feet during the motion, which is known as weight shift in golf. As pointed out by various professional players and coaches, a proper weight shift is critical in hitting a shot with good accuracy and range, and therefore it would be beneficial for golfers to obtain weight shift data corresponding to their swing motions, so that analysis and improvement on the swing pose can be made. Weight shift data collected through common methods such as using electronic scales may contain noise data due to factors such as pre-swing movements, and in order for the data to be useful, it is necessary to distinguish actual swing motion from noise. In this paper a data mining approach named Simplified Swarm Optimization with Sorted Local Search (SSO-SLS), which is based on a variant of Particle Swarm Optimization (PSO), has been proposed to classify golf swing from weight shift data. In the proposed approach a novel Sorted Local Search strategy has been introduced to remedy the issue of premature convergence facing PSO by allowing particles to obtain information from their nearest neighbors and improve swarm diversity. Experiments on UCI datasets and weight shift data in golf show that SSO-SLS is competitive with common classification techniques, and is an ideal approach for classifying golf swing from weight shift.